CN116611884A - Product recommendation method and system based on multidimensional different-composition neural network - Google Patents
Product recommendation method and system based on multidimensional different-composition neural network Download PDFInfo
- Publication number
- CN116611884A CN116611884A CN202310372655.0A CN202310372655A CN116611884A CN 116611884 A CN116611884 A CN 116611884A CN 202310372655 A CN202310372655 A CN 202310372655A CN 116611884 A CN116611884 A CN 116611884A
- Authority
- CN
- China
- Prior art keywords
- node
- nodes
- product recommendation
- product
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 31
- 230000002776 aggregation Effects 0.000 claims abstract description 34
- 238000004220 aggregation Methods 0.000 claims abstract description 34
- 230000004927 fusion Effects 0.000 claims abstract description 23
- 238000005070 sampling Methods 0.000 claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 14
- 230000002159 abnormal effect Effects 0.000 claims abstract description 9
- 230000007246 mechanism Effects 0.000 claims description 13
- 238000000605 extraction Methods 0.000 claims description 12
- 238000010276 construction Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 230000002457 bidirectional effect Effects 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 8
- 238000000059 patterning Methods 0.000 claims description 6
- 230000000694 effects Effects 0.000 description 6
- 238000005295 random walk Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000009825 accumulation Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process 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
Classifications
-
- 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
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/251—Fusion techniques of input or preprocessed data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Accounting & Taxation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application provides a product recommendation method and a system based on a multidimensional different-composition neural network, which belongs to the technical field of product recommendation, and the method comprises the following steps: step S10, acquiring user behavior data, user tag data, user social data and product attribute data, and fusing the user behavior data, the user tag data, the user social data and the product attribute data to construct an abnormal composition; step S20, sampling neighbor nodes of each node in the heterograms to obtain related neighbor nodes; step S30, carrying out double-layer aggregation on each node in the heterogram based on the related neighbor nodes to obtain fusion nodes; step S40, creating a product recommendation model based on the neural network, and training the product recommendation model by utilizing the fusion node; and S50, recommending the product by using the trained product recommendation model. The application has the advantages that: the product recommendation accuracy is greatly improved.
Description
Technical Field
The application relates to the technical field of product recommendation, in particular to a product recommendation method and system based on a multidimensional different-composition neural network.
Background
Along with the continuous expansion of the business of operators, more and more products are derived, and how to accurately find potential customers for each product becomes a problem to be solved at present. Aiming at product recommendation, a full-coverage recommendation method is adopted conventionally to recommend all products to a user, but the method has the advantages of large investment, low success rate and easiness in causing user dislike. With the development of artificial intelligence technology, the following several recommendation methods are presented:
1. respectively constructing a potential passenger identification model of each product, generating a probability value that a customer is willing to order each product, and accurately recommending to a user with high ordering probability; however, this method requires building a model based on each product, is labor intensive, and wastes computational resources; 2. according to the historical ordering behavior of the clients, a product recommendation model is built based on a collaborative filtering algorithm, and the interesting products are recommended to the users; although the method can finish the recommendation of the total products by only constructing one model, the method also has the problems of poor cold start and recommendation effect; 3. according to the method, the problem of cold start is solved, the recommending effect is improved to a certain extent, but the information acquired based on the homogeneous image is less, and the recommending effect (recommending accuracy) is still not satisfactory.
Therefore, how to provide a product recommendation method and system based on a multidimensional different-composition neural network, so as to achieve improvement of product recommendation accuracy, and the method and system are technical problems to be solved urgently.
Disclosure of Invention
The application aims to solve the technical problem of providing a product recommendation method and a system based on a multidimensional different-composition neural network, which can improve the accuracy of product recommendation.
In a first aspect, the present application provides a product recommendation method based on a multidimensional iso-patterning neural network, comprising the steps of:
step S10, acquiring user behavior data, user tag data, user social data and product attribute data, and fusing the user behavior data, the user tag data, the user social data and the product attribute data to construct an abnormal composition;
step S20, sampling neighbor nodes of each node in the heterograms to obtain related neighbor nodes;
step S30, carrying out double-layer aggregation on each node in the heterogram based on the related neighbor nodes to obtain fusion nodes;
step S40, creating a product recommendation model based on the neural network, and training the product recommendation model by utilizing the fusion node;
and S50, recommending the product by using the trained product recommendation model.
Further, in the step S10, the heterogeneous graph is composed of a plurality of nodes, and the node types of the nodes include a behavior node, a tag node, a social node and an attribute node.
Further, the step S20 specifically includes:
step S21, setting a node set, a probability p, a traversing number threshold and an upper limit of the collection number of each node type;
step S22, randomly traversing each node in the heterograms, wherein in the traversing process, the nodes walk to the neighbor node or return to the starting node based on the probability p until the number of traversed nodes reaches the threshold value of the traversing number;
step S23, judging whether the number of the corresponding node types in the node set reaches the upper limit of the collection number, if not, adding the traversed node into the node set; if yes, not adding the traversed node into a node set;
and S24, grouping the nodes in the node set based on the node type, ordering the nodes in each group in a descending order based on the occurrence times, and selecting the first k nodes as related neighbor nodes of the starting node.
Further, the step S30 specifically includes:
step S31, extracting heterogeneous information from each node;
s32, projecting the heterogeneous information into a unified space through a feature conversion formula to obtain node embedded features;
step S33, based on the node embedded characteristics, performing first layer aggregation of bidirectional LSTM on the related neighbor nodes;
and step S34, performing second-layer aggregation on the related neighbor nodes based on an attention mechanism to obtain fusion nodes.
Further, in the step S31, the extraction formula of the heterogeneous information is:
wherein x is v Heterogeneous information representing node v; d, d f Representing the dimension in which the node is embedded; r represents a set of nodes.
In a second aspect, the present application provides a product recommendation system based on a multidimensional iso-patterning neural network, including the following modules:
the system comprises a different composition construction module, a different composition generation module and a different composition generation module, wherein the different composition construction module is used for acquiring user behavior data, user tag data, user social data and product attribute data, and fusing the user behavior data, the user tag data, the user social data and the product attribute data to construct a different composition;
the node sampling module is used for sampling neighbor nodes of each node in the heterograms to obtain related neighbor nodes;
the node double-layer aggregation module is used for carrying out double-layer aggregation on each node in the heterograms based on the related neighbor nodes to obtain fusion nodes;
the product recommendation model training module is used for creating a product recommendation model based on the neural network and training the product recommendation model by utilizing the fusion node;
and the product recommendation module is used for recommending the product by using the trained product recommendation model.
Further, in the heterogeneous graph construction module, the heterogeneous graph is composed of a plurality of nodes, and the node types of the nodes comprise behavior nodes, label nodes, social nodes and attribute nodes.
Further, the node sampling module specifically includes:
the sampling parameter setting unit is used for setting a node set, a probability p, a traversal number threshold value and an upper collection number limit of each node type;
the random traversal unit is used for carrying out random traversal on each node in the abnormal composition, and in the traversal process, the nodes walk to the neighbor node or return to the starting node based on the probability p until the number of traversed nodes reaches the threshold value of the number of traversed nodes;
the node collecting unit is used for judging whether the number of the corresponding node types in the node set reaches the upper limit of the collection number, if not, the traversed node is added into the node set; if yes, not adding the traversed node into a node set;
the relevant neighbor node selection unit is used for grouping the nodes in the node set based on the node type, ordering the nodes in the groups in a descending order based on the occurrence number, and selecting the first k nodes as relevant neighbor nodes of the initial node.
Further, the node double-layer aggregation module specifically includes:
the heterogeneous information extraction unit is used for extracting heterogeneous information from each node;
the node embedded feature acquisition unit is used for projecting the heterogeneous information into a unified space through a feature conversion formula to obtain node embedded features;
the first aggregation unit is used for carrying out first layer aggregation of the bidirectional LSTM on the related neighbor nodes based on the node embedded characteristics;
and the second aggregation unit is used for carrying out second-layer aggregation on the related neighbor nodes based on the attention mechanism to obtain fusion nodes.
Further, in the heterogeneous information extraction unit, the extraction formula of the heterogeneous information is:
wherein x is v Heterogeneous information representing node v; d, d f Representing the dimension in which the node is embedded; r represents a set of nodes.
The application has the advantages that:
the different composition is constructed through four dimensions of user behavior data, user tag data, user social data and product attribute data, so that a product recommendation model can learn higher-order node information and related information, each node in the different composition is randomly traversed, in the traversing process, the nodes walk to a neighbor node or return to a starting node based on probability p until the number of traversed nodes reaches a set traversing number threshold, namely, a traditional random walk sampling mode is improved, nodes of each node type can be traversed and collected, and nodes of different node types are endowed with different weights through double-layer aggregation of the nodes, namely, attention mechanisms are introduced, so that more reasonable fusion nodes (node representation) are obtained, and finally, the product recommendation accuracy is greatly improved.
Drawings
The application will be further described with reference to examples of embodiments with reference to the accompanying drawings.
Fig. 1 is a flowchart of a product recommendation method based on a multidimensional iso-patterning neural network according to the present application.
Fig. 2 is a schematic structural diagram of a product recommendation system based on a multidimensional different-composition neural network.
FIG. 3 is a schematic diagram of the application for sampling neighbor nodes based on improved random walk.
Detailed Description
The technical scheme in the embodiment of the application has the following overall thought: the different-level composition is constructed through four dimensions, so that a product recommendation model can learn higher-level information, each node in the different-level composition is randomly traversed, in the traversing process, the nodes walk to a neighbor node or return to a starting node based on probability p until the number of traversed nodes reaches a set traversing number threshold, nodes of each node type can be traversed and collected, and a more reasonable fusion node is obtained by introducing an attention mechanism during node aggregation, so that the product recommendation accuracy is improved.
Referring to fig. 1 to 3, a preferred embodiment of a product recommendation method based on a multidimensional different-composition neural network according to the present application includes the following steps:
step S10, acquiring user behavior data, user tag data, user social data and product attribute data, and fusing the user behavior data, the user tag data, the user social data and the product attribute data to construct an abnormal composition;
if the heterogeneous graph is constructed based on the two graphs of the user product order, the heterogeneous graph can not be mined more deeply, and therefore, the heterogeneous graph is constructed through four dimensions, so that the product recommendation model can learn higher-order node information and related information, and the purpose of improving the model performance is achieved;
user behavior data, namely a common product ordering bipartite graph, takes a user and a product as nodes, and if the user historically orders a certain product, a neighbor relation is constructed between the user and the product; the heterogeneous graph constructed based on the user behavior data is often sparse, so that the product attribute data is introduced to construct an attribute relationship, for example, if both the product A and the product B have the attribute fea1_a, the node fea1_a can be introduced, and the attribute neighbor relationship is constructed among the product A, the product B and the fea1_a; the user label data is similar to the product attribute data, in the past data accumulation process, the user is marked with various labels, various user label nodes can be introduced based on the label data, and when the user is marked with the same label, a label neighbor relation is constructed among the user A, the user B and the label_a; finally, enriching neighbor relations among users through user social data, namely, after invalid call relations such as marketing, takeaway, express delivery and the like are filtered through collecting user call data, establishing the neighbor relations among the users based on a layer of call network of the users in a month, and establishing the neighbor relations among the users when call records exist among the users;
step S20, sampling neighbor nodes of each node in the heterograms to obtain related neighbor nodes;
step S30, carrying out double-layer aggregation on each node in the heterograms based on the related neighbor nodes to obtain fusion nodes, namely introducing an attention mechanism, and endowing neighbor nodes of different node types with different weights, so that better node representation is obtained, and the aim of improving the performance of a product recommendation model is fulfilled;
step S40, creating a product recommendation model based on the neural network, and training the product recommendation model by utilizing the fusion node;
and S50, recommending the product by using the trained product recommendation model.
In the step S10, the heterogeneous graph is composed of a plurality of nodes, and the node types of the nodes include a behavior node, a tag node, a social node and an attribute node.
The step S20 specifically includes:
step S21, setting a node set, a probability p, a traversing number threshold and an upper limit of the collection number of each node type; by defining the upper limit of the collection number, it is ensured that all nodes of the node type are sampled;
step S22, randomly traversing each node in the heterograms, wherein in the traversing process, the nodes walk to the neighbor node or return to the starting node based on the probability p until the number of traversed nodes reaches the threshold value of the traversing number;
step S23, judging whether the number of the corresponding node types in the node set reaches the upper limit of the collection number, if not, adding the traversed node into the node set; if yes, not adding the traversed node into a node set;
and S24, grouping the nodes in the node set based on the node type, ordering the nodes in each group in a descending order based on the occurrence times, and selecting the first k nodes as related neighbor nodes of the starting node, wherein k is a positive integer.
The node neighbor sampling method aiming at the abnormal composition is provided by the application because the traditional node neighbor sampling method based on random walk is mainly applicable to the homogeneous graph of the same node type and edge.
The step S30 specifically includes:
step S31, extracting heterogeneous information from each node;
s32, projecting the heterogeneous information into a unified space through a feature conversion formula to obtain node embedded features;
the feature conversion formula is as follows:
wherein,,d f representing the dimension in which the node is embedded; r represents a node set; x is x v Heterogeneous information representing node v; />Representing a parameter θ x FC neural network of (a);
step S33, based on the node embedded characteristics, performing first layer aggregation of bidirectional LSTM on the related neighbor nodes;
considering that each node has heterogeneous neighbors of multiple node types and each type t also has multiple heterogeneous neighbors, the nodes of the same node type are aggregated by adopting a bidirectional LSTM (least squares) so thatObtaining the type embedded f 2 (t) has a higher expression ability, type-embedding f 2 The formula (t) is specifically as follows:
wherein,,indicating the connection operation +_>And->Respectively representing forward and reverse LSTM;
step S34, performing second-layer aggregation on the related neighbor nodes based on an attention mechanism to obtain fusion nodes; attention mechanisms are introduced because the contributions of related neighbor nodes of different node types are different.
Node v i The final embedded computation of (2) is expressed as:
wherein T is a set of types comprising heterogeneous nodes;the impact value representing the generation of each node type on node v is calculated as follows:
wherein, the LeakyReLU represents an activation function; u epsilon R 2d*l Representing an attention parameter; f (F) 2 (T) representation type embedding f 2 A collection of (t); t e T, representing the type of a heterogeneous node;
it is noted that for node v, the same dimension d is used in the transformation node feature embedding, the aggregate t-type neighborhood embedding, and the combined final embedding, making model tuning easier;
in the step S31, the extraction formula of the heterogeneous information is:
wherein x is v Heterogeneous information representing node v; d, d f Representing the dimension in which the node is embedded; r represents a set of nodes.
The application relates to a product recommendation method based on a multidimensional different-composition neural network, which comprises the following modules:
the system comprises a different composition construction module, a different composition generation module and a different composition generation module, wherein the different composition construction module is used for acquiring user behavior data, user tag data, user social data and product attribute data, and fusing the user behavior data, the user tag data, the user social data and the product attribute data to construct a different composition;
if the heterogeneous graph is constructed based on the two graphs of the user product order, the heterogeneous graph can not be mined more deeply, and therefore, the heterogeneous graph is constructed through four dimensions, so that the product recommendation model can learn higher-order node information and related information, and the purpose of improving the model performance is achieved;
user behavior data, namely a common product ordering bipartite graph, takes a user and a product as nodes, and if the user historically orders a certain product, a neighbor relation is constructed between the user and the product; the heterogeneous graph constructed based on the user behavior data is often sparse, so that the product attribute data is introduced to construct an attribute relationship, for example, if both the product A and the product B have the attribute fea1_a, the node fea1_a can be introduced, and the attribute neighbor relationship is constructed among the product A, the product B and the fea1_a; the user label data is similar to the product attribute data, in the past data accumulation process, the user is marked with various labels, various user label nodes can be introduced based on the label data, and when the user is marked with the same label, a label neighbor relation is constructed among the user A, the user B and the label_a; finally, enriching neighbor relations among users through user social data, namely, after invalid call relations such as marketing, takeaway, express delivery and the like are filtered through collecting user call data, establishing the neighbor relations among the users based on a layer of call network of the users in a month, and establishing the neighbor relations among the users when call records exist among the users;
the node sampling module is used for sampling neighbor nodes of each node in the heterograms to obtain related neighbor nodes;
the node double-layer aggregation module is used for carrying out double-layer aggregation on each node in the heterograms based on the related neighbor nodes to obtain fusion nodes, namely introducing an attention mechanism, and endowing neighbor nodes of different node types with different weights, so that better node representation is obtained, and the aim of improving the performance of a product recommendation model is fulfilled;
the product recommendation model training module is used for creating a product recommendation model based on the neural network and training the product recommendation model by utilizing the fusion node;
and the product recommendation module is used for recommending the product by using the trained product recommendation model.
In the heterogeneous graph construction module, the heterogeneous graph is composed of a plurality of nodes, and the node types of the nodes comprise behavior nodes, label nodes, social nodes and attribute nodes.
The node sampling module specifically comprises:
the sampling parameter setting unit is used for setting a node set, a probability p, a traversal number threshold value and an upper collection number limit of each node type; by defining the upper limit of the collection number, it is ensured that all nodes of the node type are sampled;
the random traversal unit is used for carrying out random traversal on each node in the abnormal composition, and in the traversal process, the nodes walk to the neighbor node or return to the starting node based on the probability p until the number of traversed nodes reaches the threshold value of the number of traversed nodes;
the node collecting unit is used for judging whether the number of the corresponding node types in the node set reaches the upper limit of the collection number, if not, the traversed node is added into the node set; if yes, not adding the traversed node into a node set;
the relevant neighbor node selection unit is used for grouping all nodes in the node set based on node types, ordering the nodes in all the groups in a descending order based on the occurrence times, selecting the first k nodes as relevant neighbor nodes of the starting node, and k is a positive integer.
The node neighbor sampling method aiming at the abnormal composition is provided by the application because the traditional node neighbor sampling method based on random walk is mainly applicable to the homogeneous graph of the same node type and edge.
The node double-layer aggregation module specifically comprises:
the heterogeneous information extraction unit is used for extracting heterogeneous information from each node;
the node embedded feature acquisition unit is used for projecting the heterogeneous information into a unified space through a feature conversion formula to obtain node embedded features;
the feature conversion formula is as follows:
wherein,,d f representing the dimension in which the node is embedded; r represents a node set; x is x v Heterogeneous information representing node v; />Representing a parameter θ x FC neural network of (a);
the first aggregation unit is used for carrying out first layer aggregation of the bidirectional LSTM on the related neighbor nodes based on the node embedded characteristics;
considering that each node has heterogeneous neighbors of multiple node types and each type t also has multiple heterogeneous neighbors, the nodes of the same node type are aggregated by adopting a bidirectional LSTM so that the type embedding f is learned 2 (t) has a higher expression ability, type-embedding f 2 The formula (t) is specifically as follows:
wherein,,indicating the connection operation +_>And->Respectively representing forward and reverse LSTM;
the second aggregation unit is used for carrying out second-layer aggregation on the related neighbor nodes based on an attention mechanism to obtain fusion nodes; attention mechanisms are introduced because the contributions of related neighbor nodes of different node types are different.
Node v i The final embedded computation of (2) is expressed as:
wherein T is a set of types comprising heterogeneous nodes;the impact value representing the generation of each node type on node v is calculated as follows:
wherein, the LeakyReLU represents an activation function; u epsilon R 2d*l Representing an attention parameter; f (F) 2 (T) representation type embedding f 2 A collection of (t); t e T, representing the type of a heterogeneous node;
it is noted that for node v, the same dimension d is used in the transformation node feature embedding, the aggregate t-type neighborhood embedding, and the combined final embedding, making model tuning easier;
in the heterogeneous information extraction unit, the extraction formula of the heterogeneous information is as follows:
wherein x is v Heterogeneous information representing node v; d, d f Representing the dimension in which the node is embedded; r represents a set of nodes.
And evaluating the effect of the product recommendation model in an offline mode, specifically splitting the original data into a training set and a test set, training the model by using the training set, and evaluating the model effect by using the test set. Based on the application scene, the data before the user T month is used as a training set, the data of the user T month is used as a test set, and the accuracy rate and the recall rate are used for evaluating the final recommendation effect.
Suppose that the candidate set predicted to be recommended to the user by the product recommendation model is R u (N) the product candidate set truly ordered by the user is A u The total number of users recommended by the product recommendation model is set U, where N is the recommended number of products, and for user U, the final accuracy P u And recall rate R u Is defined as:
average accuracy MP of all users in a test set u And average recall MR u The definition is as follows:
finally, compared with the traditional collaborative filtering recommendation algorithm, the average accuracy and average recall rate of the method on the test set are respectively improved by 10.1% and 8.9%.
In summary, the application has the advantages that:
the different composition is constructed through four dimensions of user behavior data, user tag data, user social data and product attribute data, so that a product recommendation model can learn higher-order node information and related information, each node in the different composition is randomly traversed, in the traversing process, the nodes walk to a neighbor node or return to a starting node based on probability p until the number of traversed nodes reaches a set traversing number threshold, namely, a traditional random walk sampling mode is improved, nodes of each node type can be traversed and collected, and nodes of different node types are endowed with different weights through double-layer aggregation of the nodes, namely, attention mechanisms are introduced, so that more reasonable fusion nodes (node representation) are obtained, and finally, the product recommendation accuracy is greatly improved.
While specific embodiments of the application have been described above, it will be appreciated by those skilled in the art that the specific embodiments described are illustrative only and not intended to limit the scope of the application, and that equivalent modifications and variations of the application in light of the spirit of the application will be covered by the claims of the present application.
Claims (10)
1. A product recommendation method based on a multidimensional different-composition neural network is characterized by comprising the following steps of: the method comprises the following steps:
step S10, acquiring user behavior data, user tag data, user social data and product attribute data, and fusing the user behavior data, the user tag data, the user social data and the product attribute data to construct an abnormal composition;
step S20, sampling neighbor nodes of each node in the heterograms to obtain related neighbor nodes;
step S30, carrying out double-layer aggregation on each node in the heterogram based on the related neighbor nodes to obtain fusion nodes;
step S40, creating a product recommendation model based on the neural network, and training the product recommendation model by utilizing the fusion node;
and S50, recommending the product by using the trained product recommendation model.
2. The product recommendation method based on the multidimensional iso-patterning neural network as claimed in claim 1, wherein: in the step S10, the heterogeneous graph is composed of a plurality of nodes, and the node types of the nodes include a behavior node, a tag node, a social node and an attribute node.
3. The product recommendation method based on the multidimensional iso-patterning neural network as claimed in claim 1, wherein: the step S20 specifically includes:
step S21, setting a node set, a probability p, a traversing number threshold and an upper limit of the collection number of each node type;
step S22, randomly traversing each node in the heterograms, wherein in the traversing process, the nodes walk to the neighbor node or return to the starting node based on the probability p until the number of traversed nodes reaches the threshold value of the traversing number;
step S23, judging whether the number of the corresponding node types in the node set reaches the upper limit of the collection number, if not, adding the traversed node into the node set; if yes, not adding the traversed node into a node set;
and S24, grouping the nodes in the node set based on the node type, ordering the nodes in each group in a descending order based on the occurrence times, and selecting the first k nodes as related neighbor nodes of the starting node.
4. The product recommendation method based on the multidimensional iso-patterning neural network as claimed in claim 1, wherein: the step S30 specifically includes:
step S31, extracting heterogeneous information from each node;
s32, projecting the heterogeneous information into a unified space through a feature conversion formula to obtain node embedded features;
step S33, based on the node embedded characteristics, performing first layer aggregation of bidirectional LSTM on the related neighbor nodes;
and step S34, performing second-layer aggregation on the related neighbor nodes based on an attention mechanism to obtain fusion nodes.
5. The method for recommending products based on multidimensional heterograph neural network of claim 4, wherein: in the step S31, the extraction formula of the heterogeneous information is:
wherein x is v Heterogeneous information representing node v; d, d f Representing the dimension in which the node is embedded; r represents a set of nodes.
6. A product recommendation system based on a multidimensional different composition neural network is characterized in that: the device comprises the following modules:
the system comprises a different composition construction module, a different composition generation module and a different composition generation module, wherein the different composition construction module is used for acquiring user behavior data, user tag data, user social data and product attribute data, and fusing the user behavior data, the user tag data, the user social data and the product attribute data to construct a different composition;
the node sampling module is used for sampling neighbor nodes of each node in the heterograms to obtain related neighbor nodes;
the node double-layer aggregation module is used for carrying out double-layer aggregation on each node in the heterograms based on the related neighbor nodes to obtain fusion nodes;
the product recommendation model training module is used for creating a product recommendation model based on the neural network and training the product recommendation model by utilizing the fusion node;
and the product recommendation module is used for recommending the product by using the trained product recommendation model.
7. The multi-dimensional heterographic neural network based product recommendation system of claim 6, wherein: in the heterogeneous graph construction module, the heterogeneous graph is composed of a plurality of nodes, and the node types of the nodes comprise behavior nodes, label nodes, social nodes and attribute nodes.
8. The multi-dimensional heterographic neural network based product recommendation system of claim 6, wherein: the node sampling module specifically comprises:
the sampling parameter setting unit is used for setting a node set, a probability p, a traversal number threshold value and an upper collection number limit of each node type;
the random traversal unit is used for carrying out random traversal on each node in the abnormal composition, and in the traversal process, the nodes walk to the neighbor node or return to the starting node based on the probability p until the number of traversed nodes reaches the threshold value of the number of traversed nodes;
the node collecting unit is used for judging whether the number of the corresponding node types in the node set reaches the upper limit of the collection number, if not, the traversed node is added into the node set; if yes, not adding the traversed node into a node set;
the relevant neighbor node selection unit is used for grouping the nodes in the node set based on the node type, ordering the nodes in the groups in a descending order based on the occurrence number, and selecting the first k nodes as relevant neighbor nodes of the initial node.
9. The multi-dimensional heterographic neural network based product recommendation system of claim 6, wherein: the node double-layer aggregation module specifically comprises:
the heterogeneous information extraction unit is used for extracting heterogeneous information from each node;
the node embedded feature acquisition unit is used for projecting the heterogeneous information into a unified space through a feature conversion formula to obtain node embedded features;
the first aggregation unit is used for carrying out first layer aggregation of the bidirectional LSTM on the related neighbor nodes based on the node embedded characteristics;
and the second aggregation unit is used for carrying out second-layer aggregation on the related neighbor nodes based on the attention mechanism to obtain fusion nodes.
10. The multi-dimensional heterographic neural network based product recommendation system of claim 9, wherein: in the heterogeneous information extraction unit, the extraction formula of the heterogeneous information is as follows:
wherein x is v Heterogeneous information representing node v; d, d f Representing the dimension in which the node is embedded; r represents a set of nodes.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310372655.0A CN116611884A (en) | 2023-04-10 | 2023-04-10 | Product recommendation method and system based on multidimensional different-composition neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310372655.0A CN116611884A (en) | 2023-04-10 | 2023-04-10 | Product recommendation method and system based on multidimensional different-composition neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116611884A true CN116611884A (en) | 2023-08-18 |
Family
ID=87682496
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310372655.0A Pending CN116611884A (en) | 2023-04-10 | 2023-04-10 | Product recommendation method and system based on multidimensional different-composition neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116611884A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117390266A (en) * | 2023-10-08 | 2024-01-12 | 宁夏大学 | Project recommendation method based on high-order neighbor generation algorithm and heterogeneous graph neural network |
CN117493490A (en) * | 2023-11-17 | 2024-02-02 | 南京信息工程大学 | Topic detection method, device, equipment and medium based on heterogeneous multi-relation graph |
-
2023
- 2023-04-10 CN CN202310372655.0A patent/CN116611884A/en active Pending
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117390266A (en) * | 2023-10-08 | 2024-01-12 | 宁夏大学 | Project recommendation method based on high-order neighbor generation algorithm and heterogeneous graph neural network |
CN117390266B (en) * | 2023-10-08 | 2024-04-30 | 宁夏大学 | Project recommendation method based on high-order neighbor generation algorithm and heterogeneous graph neural network |
CN117493490A (en) * | 2023-11-17 | 2024-02-02 | 南京信息工程大学 | Topic detection method, device, equipment and medium based on heterogeneous multi-relation graph |
CN117493490B (en) * | 2023-11-17 | 2024-05-14 | 南京信息工程大学 | Topic detection method, device, equipment and medium based on heterogeneous multi-relation graph |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN116611884A (en) | Product recommendation method and system based on multidimensional different-composition neural network | |
CN110263280B (en) | Multi-view-based dynamic link prediction depth model and application | |
CN103902538B (en) | Information recommending apparatus and method based on decision tree | |
CN112613602A (en) | Recommendation method and system based on knowledge-aware hypergraph neural network | |
CN112149352B (en) | Prediction method for marketing activity clicking by combining GBDT automatic characteristic engineering | |
CN112507224B (en) | Service recommendation method of man-machine object fusion system based on heterogeneous network representation learning | |
CN113051468B (en) | Movie recommendation method and system based on knowledge graph and reinforcement learning | |
CN114911870A (en) | Fusion management framework for multi-source heterogeneous industrial data | |
CN104778237A (en) | Individual recommending method and system based on key users | |
Mahyar et al. | Centrality-based group formation in group recommender systems | |
CN114595383A (en) | Marine environment data recommendation method and system based on session sequence | |
CN117495481B (en) | Article recommendation method based on heterogeneous timing diagram attention network | |
CN111949892A (en) | Multi-relation perception temporal interaction network prediction method | |
CN113868537B (en) | Recommendation method based on multi-behavior session graph fusion | |
He et al. | Evolutionary community detection in social networks | |
CN113610170A (en) | Influence maximization method based on time sequence network community detection | |
CN117440182B (en) | Intelligent recommendation method and system based on video content analysis and user labels | |
CN112907056B (en) | Urban management complaint event prediction method and system based on graph neural network | |
CN117422134A (en) | Knowledge graph recommendation method based on graph convolution neural network | |
CN112016701A (en) | Abnormal change detection method and system integrating time sequence and attribute behaviors | |
CN114996584B (en) | Diversity perception interaction recommendation method based on deep reinforcement learning | |
CN116304336A (en) | Course recommendation method integrating knowledge graph and graph neural network | |
WO2016116958A1 (en) | Sequential data analysis device and program | |
CN115599990A (en) | Knowledge perception and deep reinforcement learning combined cross-domain recommendation method and system | |
CN115545833A (en) | Recommendation method and system based on user social information |
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 |