CN117994011A - E-commerce dynamic perception data recommendation method based on memory updating and neighbor transfer - Google Patents
E-commerce dynamic perception data recommendation method based on memory updating and neighbor transfer Download PDFInfo
- Publication number
- CN117994011A CN117994011A CN202410406224.6A CN202410406224A CN117994011A CN 117994011 A CN117994011 A CN 117994011A CN 202410406224 A CN202410406224 A CN 202410406224A CN 117994011 A CN117994011 A CN 117994011A
- Authority
- CN
- China
- Prior art keywords
- node
- commodity
- user
- neighbor
- information
- 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 87
- 230000015654 memory Effects 0.000 title claims abstract description 75
- 230000008447 perception Effects 0.000 title claims abstract description 44
- 238000012546 transfer Methods 0.000 title claims abstract description 23
- 239000013598 vector Substances 0.000 claims abstract description 35
- 230000006399 behavior Effects 0.000 claims abstract description 9
- 238000011156 evaluation Methods 0.000 claims abstract description 9
- 230000004931 aggregating effect Effects 0.000 claims abstract description 5
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 30
- 230000008569 process Effects 0.000 claims description 30
- 230000003993 interaction Effects 0.000 claims description 28
- 238000004364 calculation method Methods 0.000 claims description 14
- 230000002776 aggregation Effects 0.000 claims description 12
- 238000004220 aggregation Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 238000003860 storage Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 5
- 238000004140 cleaning Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 description 13
- 238000013528 artificial neural network Methods 0.000 description 12
- 238000013135 deep learning Methods 0.000 description 10
- 230000007246 mechanism Effects 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 7
- 230000005540 biological transmission Effects 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- TUSUWHFYKZZRIG-JQWMYKLHSA-N C([C@@H](NC(=O)[C@@H](C(C)C)NC(=O)[C@@H](CC(C)C)NC)C(=O)N[C@H](CC=1C=CC=CC=1)C(=O)N[C@H](CC(C)C)C(N)=O)C1=CC=CC=C1 Chemical compound C([C@@H](NC(=O)[C@@H](C(C)C)NC(=O)[C@@H](CC(C)C)NC)C(=O)N[C@H](CC=1C=CC=CC=1)C(=O)N[C@H](CC(C)C)C(N)=O)C1=CC=CC=C1 TUSUWHFYKZZRIG-JQWMYKLHSA-N 0.000 description 4
- 238000002679 ablation Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000001914 filtration Methods 0.000 description 4
- 230000000306 recurrent effect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 206010044565 Tremor Diseases 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006854 communication Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000004941 influx Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000000463 material Substances 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
- 238000005457 optimization Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012360 testing method Methods 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
- G06F18/10—Pre-processing; Data cleansing
-
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an e-commerce dynamic perception data recommendation method based on memory updating and neighbor transfer, which belongs to the technical field of e-commerce recommendation and comprises the following steps: step 1, acquiring commodity information, price, sales volume, evaluation and user behavior data from an electronic commerce platform, constructing an electronic commerce dynamic perception data set and preprocessing to obtain a time sequence data set; step 2, inputting the time sequence data set into a memory updating module for capturing node information, aggregating information and updating memory; step 3, neighbor information characteristics are acquired based on a neighbor transfer module to generate neighbor characteristics of the node vector; step 4, generating node embedding based on the multi-head attention module; and 5, decoding the obtained node embedding by adopting a multi-layer perceptron to obtain the commodity which is recommended to be interested by the user and recommending the commodity. The method and the device can efficiently acquire the neighbor information characteristics and more accurately recommend the possibly interested commodity to the user.
Description
Technical Field
The invention belongs to the technical field of e-commerce recommendation, and particularly relates to an e-commerce dynamic perception data recommendation method based on memory updating and neighbor transfer.
Background
With the popularization of the internet, the electric business users are faced with the influx of a large amount of information, and the demands of the electric business users for the information are different in different time and context, so that the traditional static list or search result cannot meet the demands of the electric business users, how to deeply analyze and mine the dynamic perception data becomes a cross research hotspot of a recommendation system and data mining, and the research method of the recommendation system and the dynamic data mining is approximately as follows:
Study of the recommendation algorithm on data prediction: in order to convert "user interests" into "personalized recommendations", statistical analysis of user behavior data is a key element in the recommendation system. There has been considerable research in the field of processing and prediction of recommendation algorithms both domestically and abroad, and early methods used collaborative filtering-based methods, such as collaborative filtering algorithms, to analyze the interaction behavior between users and items to provide recommendations. Recent studies have increasingly used machine learning methods to predict user interests, such as collaborative filtering algorithms that incorporate user features and content information, so that e-commerce websites can more accurately predict the range of user purchase interests, thereby providing personalized recommendations. For example, based on historical click and purchase data of a user, a user interest prediction model is established, and deep learning methods such as Neural Collaborative Filtering (NCF) are used to predict items of likely interest to the user. The method can track the interest trend of the user and improve the effect of personalized recommendation.
Study on dynamic map data prediction study: methods for predicting dynamic map data can be generally divided into two main categories: firstly, a traditional graph algorithm and time sequence modeling method are adopted; secondly, a method based on deep learning, in particular to a deep learning technology suitable for graph data. The traditional method comprises the following steps: early dynamic graph data prediction methods relied primarily on graph-based algorithms and time series models. One such method is to use time series modeling, such as the autoregressive integrated moving average (ARIMA) method, to capture periodicity and trending information in the dynamic map. Such conventional models have a high degree of prediction accuracy in processing time series data. The deep learning method comprises the following steps: recent studies increasingly employ deep learning methods to address the problem of prediction of dynamic image data. Among other things, processing dynamic graph data using deep learning techniques such as Graph Neural Networks (GNNs) and graph roll-up neural networks (GCNs) has met with significant success. These methods can take into account the variation of nodes and edges over different time steps to better capture complex relationships in the dynamic graph. For example, researchers have applied deep learning methods to colliery downhole face pressure predictions. They used modified Recurrent Neural Networks (RNNs) such as long and short term memory networks (LSTM) to build a prediction model of mine pressure, accounting for the variation in face mine pressure over different time steps. Experimental results show that the LSTM method has higher accuracy compared with the traditional neural network.
From the above research, the deep learning method is better achieved in the e-commerce dynamic perception data prediction, but in practical application, the single prediction method cannot well capture the high-order characteristics of the e-commerce dynamic perception data. Therefore, by deeply analyzing the characteristics of the dynamic time series data, the predictive analysis by applying the composite model fused by different dynamic analysis methods is a trend of solving the problem of the prediction of the dynamic time series data of the electronic commerce in the future.
The prediction accuracy of the dynamic perception data in the prior art is not very high, and due to the complex dependence of the dynamic graph model, the defects of excessive smoothness and excessive time consumption can be overcome when the number of layers is too high in the process of inquiring the graph. Meanwhile, models such as LSTM, GRU and the like are generally used for solving the problem of long-term dependence commonly existing in a general recurrent neural network, and in the process suitable for a dynamic graph neural network, events cannot be ensured to arrive according to the sequence of time stamps, so that the performance of the model is unstable.
Disclosure of Invention
In order to solve the problems, the invention provides an electronic commerce dynamic perception data recommendation method based on memory updating and neighbor delivery, which adopts a memory updating strategy, a time stamp coding strategy, a neighbor delivery strategy, a multi-head attention mechanism and the like to efficiently acquire neighbor information characteristics, screens important information, improves the execution efficiency and the information extraction capability of a model, and further more accurately recommends possibly interested commodities to a user.
The technical scheme of the invention is as follows:
A method for recommending e-commerce dynamic perception data based on memory updating and neighbor transfer comprises the following steps:
Step 1, acquiring commodity information, price, sales volume, evaluation and user behavior data from an electronic commerce platform, constructing an electronic commerce dynamic perception data set and preprocessing to obtain a time sequence data set;
step 2, inputting the time sequence data set into a memory updating module for capturing node information, aggregating information and updating memory;
Step 3, neighbor information characteristics are acquired based on a neighbor transfer module to generate neighbor characteristics of the node vector;
step 4, generating node embedding based on the multi-head attention module;
and 5, decoding the obtained node embedding by adopting a multi-layer perceptron to obtain the commodity which is recommended to be interested by the user and recommending the commodity.
Further, the specific process of the step 1 is as follows:
step 1.1, acquiring commodity information, price, sales volume, evaluation and user behavior data through an API interface provided by an electronic commerce platform, and constructing and obtaining an electronic commerce dynamic perception data set from the data E-commerce dynamic perception dataset/>Is a graph network structure; wherein/>Representing e-commerce dynamic perception dataset/>User node in/>A user node sequence number; Representing e-commerce dynamic perception dataset/> Commodity node in/>The commodity node sequence number; /(I)Representing the current moment of the user commodity interaction event; /(I)Representing a duration of a user merchandise interaction event; /(I)Dynamic perception of datasets/>, for e-commerceMedium user node/>With commodity node/>Edges connected with each other; /(I)Representing the total number of user nodes,/>Representing the total number of commodity nodes;
Step 1.2, dynamically sensing data set for E-commerce Cleaning, and filling the missing values by using a linear interpolation method;
Step 1.3, extracting and sequencing data by utilizing numpy packets in python, storing edge characteristics, and finally obtaining a time sequence data set after digital signal processing ; Wherein the time series dataset/>Is a graph network structure; /(I)For time series dataset/>User nodes in (a); /(I)For time series dataset/>Commodity nodes in (a); /(I)Representing a time series dataset/>Medium user node/>With commodity node/>Edges connected with each other and corresponding to the current time/>User commodity interaction event.
Further, the specific process of the step 2 is as follows:
Step 2.1, acquiring initial information of the node through a node information capturing unit, wherein a calculation formula is as follows:
(1)；
(2)；
Wherein, 、/>User nodes/>, respectivelyCommodity node/>At the present moment/>A calculated message; a multi-layer perceptron for a learnable message function; /(I) Representing the last moment of the commodity interaction event of the user;、/> User nodes/>, respectively Commodity node/>A memory vector of the previous moment;
Step 2.2, the aggregation of the user node information and the commodity node information is realized through an information aggregation unit, and the calculation formula is as follows:
(3)；
(4)；
Wherein, 、/>User nodes/>, respectivelyCommodity node/>An aggregated message; /(I) 、User nodes/>, respectivelyCommodity node/>At time/>A calculated message;
And 2.3, optimizing the storage of the user node and the commodity node through a memory updating unit, wherein the calculation formula is as follows:
(5)；
(6)；
Wherein, 、/>User nodes/>, respectivelyCommodity node/>Is a memory of (a); /(I)Is a learnable memory update function.
Further, the specific process of the step 3 is as follows:
and 3.1, transmitting the two-order neighbor information, wherein the calculation formula is as follows:
(7)；
(8)；
Wherein, 、/>Current moment/>, respectivelyUser node/>Commodity nodeNeighbor information of (2); /(I)Is an update function; /(I)、/>Current moment/>, respectivelyBy/>、/>Embedding the transmitted nodes;
step 3.2, carrying out vector list storage by adopting a neighbor generating function, and storing and transmitting neighbor information; the specific process is as follows:
(9)；
(10)；
Wherein, 、/>Current moment/>, respectivelyUser node/>Commodity node/>First/>Characteristic information of each neighbor node; /(I)Is the total number of neighbor nodes; /(I)To summarize the historical state functions of the neighbors of the user node or commodity node; /(I)、/>User node/>, respectively, at the last momentCommodity node/>Neighbor features of (a); /(I)Is an average strategy; /(I)As an attenuation or mapping function; /(I)Transmitting a sampling strategy for neighbor information;
Step 3.3, performing position coding, wherein the specific process is as follows:
(11)；
(12)；
(13)；
(14)；
(15)；
(16)；
Wherein, 、/>Current moment/>, respectivelyUser node/>Commodity node/>Neighbor features after position coding; /(I)、/>Current moment/>, respectivelyUser node/>Commodity node/>Information vectors of different neighbor nodes;、/> Current moment/>, respectively User node/>Commodity node/>An absolute position encoding vector; /(I)、Current moment/>, respectivelyUser node/>Commodity node/>First/>Characteristic information of each neighbor node; /(I)、/>User nodes/>, respectivelyCommodity node/>First/>Absolute positions of the neighboring nodes.
Further, the specific process of the step 4 is as follows:
User node Computing memory/>, corresponding to a message, at different timesAnd/>Input multi-head attention module generates user node embedding/>，/>For time/>User node/>Is a memory of (a); the specific formula is as follows:
(17)；
(18)；
(19)；
(20)；
Node of commodity Computing memory/>, corresponding to a message, at different timesAnd/>Input multi-head attention module for generating commodity node embedding/>，/>For time/>Commodity node/>Is a memory of (a); the specific formula is as follows:
(21)；
(22)；
(23)；
(24)；
Wherein, As a multi-headed attention function; /(I)、/>User nodes/>, respectivelyCommodity node/>Is a query vector of (1); /(I)、/>User nodes/>, respectivelyCommodity node/>Is a key vector of (a); /(I)、/>User nodes/>, respectivelyCommodity node/>Is a vector of values of (a); /(I)Is an activation function; /(I)Transpose the symbol; /(I)Is a scaling factor; /(I)、User node/>, respectively, at the last momentCommodity node/>Is a memory of (a); /(I)、/>、/>、/>、/>、Respectively/>、/>、/>、/>、/>、/>Is a learning weight matrix.
Further, the specific process of the step 5 is as follows:
step 5.1, calculating a predicted value of interest of the user to the commodity, wherein the specific formula is as follows:
(25)；
Wherein, A predicted value indicating whether or not the commodity is of interest is a predicted value of 0 to 1; /(I)Representing a nonlinear activation function; /(I)、/>Are all learnable weights; /(I)Representing a splicing operation; /(I)、/>Are both learnable biases;
Step 5.2, presetting a threshold value if And if the threshold value is greater than or equal to the threshold value, the user node/>, is representedFor commodity node/>Interest, otherwise represent user node/>For commodity node/>Not of interest; by user node/>The prediction calculation is carried out with different commodity nodes, so that the user/>Is a plurality of items of interest.
The invention has the beneficial technical effects that: aiming at the problems that the traditional dynamic graph model is easy to cause excessive smoothness and lose rich characteristic information when the number of layers is too high in query operation, the invention introduces a memory updating strategy, is beneficial to better keeping the time sequence information of the nodes, and thus, the problem of excessive smoothness is alleviated; aiming at the stability problem in the traditional dynamic graph model, particularly the problem that events cannot arrive in the time stamp sequence, the method introduces a time stamp coding strategy, and ensures that node messages are transmitted to neighbor nodes in the time stamp sequence, thereby improving the grasp of the model on the event time sequence; through the neighbor transmission strategy and the multi-head attention mechanism, the method and the device can efficiently acquire neighbor information characteristics, screen important information, improve the execution efficiency and the information extraction capability of the model, and more accurately recommend possibly interested commodities to a user.
Drawings
FIG. 1 is a flow chart of the method for recommending e-commerce dynamic perception data based on memory updating and neighbor delivery.
FIG. 2 is a graph showing the comparison of AP indicators of different models in experiment 2 according to the present invention.
FIG. 3 is a graph showing the comparison of AUC indexes of different models in experiment 2 according to the present invention.
Detailed Description
The invention is described in further detail below with reference to the attached drawings and detailed description:
With the development of 5G and artificial intelligence technology, the material living level and the consumption entertainment mode of people are changed over the sky, so that the commodities and services in the current age are rapidly increased, the interest preference of different people is quite different, and the recommendation method is generated under the drive of the large environment background. The invention provides an e-commerce dynamic perception data recommendation method based on memory updating and neighbor transmission on the basis of the method, which mainly comprises a memory updating module, a multi-head attention module and a neighbor transmission module. The method specifically comprises the following steps: the memory updating module acquires initial information of the node through a node information capturing strategy, and captures time information of the node through a node information aggregation and updating method; the neighbor transfer module firstly acquires neighbor information based on a time edge transfer strategy, and then stores the neighbor information in a parallel operation mode; the multi-head attention is operated by the multi-head attention on the basis of the memory updating module and the neighbor transmission module, and the current node embedding is created. And finally, decoding the obtained node embedding by the multi-layer perceptron, and recommending the interesting commodity to the user.
The invention provides an electronic commerce dynamic perception data recommendation method based on memory updating and neighbor transmission, which takes electronic commerce dynamic perception data recommendation as a research object and improves the prediction precision of a prediction model of the electronic commerce dynamic perception data recommendation as a core target, and specifically comprises the following steps:
And step 1, acquiring commodity information, price, sales volume, evaluation and user behavior data from an electronic commerce platform, constructing an electronic commerce dynamic perception data set and preprocessing to obtain a time sequence data set.
Step 1.1, most e-commerce platforms, such as Taobao, jindong, jian Duo, tremble, etc., provide an API interface for the developer. Through the API interfaces, developers can acquire data such as commodity information, price, sales volume, evaluation, user behavior and the like, and construct and obtain an e-commerce dynamic perception data set from the dataE-commerce dynamic perception dataset/>Is a graph network structure; wherein/>Representing e-commerce dynamic perception dataset/>User node in/>A user node sequence number; Representing e-commerce dynamic perception dataset/> Commodity node in/>The commodity node sequence number; /(I)Representing the current moment of the user commodity interaction event; /(I)Representing a duration of a user merchandise interaction event; /(I)Dynamic perception of datasets/>, for e-commerceMedium user node/>With commodity node/>Edges connected with each other; /(I)Representing the total number of user nodes,/>Representing the total number of commodity nodes; when a user clicks a commodity, interaction occurs between the user and the commodity, and the process is called a user commodity interaction event;
Step 1.2, dynamically sensing data set for E-commerce Performing simple cleaning, and filling the missing values by using a linear interpolation method;
Step 1.3, extracting and sequencing data by utilizing numpy packets in python, storing edge characteristics, and finally obtaining a time sequence data set after digital signal processing . Wherein the time series dataset/>Is a graph network structure; /(I)For time series dataset/>User nodes in (a); /(I)For time series dataset/>Commodity nodes in (a); /(I)Representing a time series dataset/>Medium user node/>With commodity node/>Edges connected with each other and corresponding to the current time/>User commodity interaction event.
The object of the present invention is to process time-stamped event sequence data, capturing time dependencies through interactions between learning nodes, for recommending items of interest to a user. The inventive design incorporates a variety of techniques including memory updating, multi-head attention, neighbor delivery, and position coding to improve representation learning and predictive performance of time-dynamic graph data.
Step 2, inputting the time sequence data set into a memory updating module for capturing node information, aggregating information and updating memory;
the memory updating module acquires initial information of the node through a node information capturing strategy, and captures time information of the node through a node information aggregation and updating method. The memory updating module adopts a memory updating strategy to obtain the time information characteristics of the nodes. The method can help the model to better reflect the time sequence characteristics of the nodes through capturing the node information, aggregating the information and updating the memory.
The memory updating module is mainly used for acquiring node time interaction information and mainly comprises three units: the system comprises a node information capturing unit, an information aggregation unit and a memory updating unit.
The memory updating module learns the node representation from the time sequence of the data and simultaneously performs efficient parallel processing. The specific process of the step 2 is as follows:
Step 2.1, acquiring initial information of a node through a node information capturing unit; the specific process is as follows:
user commodity interaction event in E-commerce environment When this happens, the involved user node/>And commodity node/>Interactions between them occur requiring the messages to be calculated and their memory to be updated. The computation of the message takes into account the interaction events between the nodes, the time stamp information and the memorization of the nodes.
For each user commodity interaction event, the involved user nodeAnd commodity node/>Two messages are calculated as follows:
(1)；
(2)；
Wherein, 、/>User nodes/>, respectivelyCommodity node/>At the present moment/>A calculated message; A multi-layer perceptron (MLPs) for a learnable message function; /(I) Representing the last moment of the commodity interaction event of the user;、/> User nodes/>, respectively Commodity node/>A memory vector of the previous moment;
Step 2.2, the aggregation of the user node information and the commodity node information is realized through an information aggregation unit; the specific process is as follows:
Because the user node may interact with the commodity node multiple times at different times, the messages are aggregated using an average aggregation mechanism to obtain time information for the user node and the commodity node.
In order to improve the calculation efficiency, the model adopts an average aggregation mechanism to aggregate the messages of a plurality of user commodity interaction events generated by the same user node or the same commodity node at the previous moment, and the calculation formula is as follows:
(3)；
(4)；
Wherein, 、/>User nodes/>, respectivelyCommodity node/>An aggregated message; /(I) 、User nodes/>, respectivelyCommodity node/>At time/>A calculated message; /(I) 、/>The time/>, respectively, is the user node 1 and the commodity node 1A calculated message;
and 2.3, optimizing the storage of the user node and the commodity node through a memory updating unit. The specific process is as follows:
(5)；
(6)；
Wherein, 、/>Current moment/>, respectivelyUser node/>Commodity node/>Is a memory of (a); is a learnable memory update function, such as a recurrent neural network LSTM, for updating memory. Memory device 、/>Updated after each event occurs, involving the user node/>And commodity node/>Is a virtual object, is an interactive event of a virtual object. Memory/>、/>The vector generated for the memory update module is interacted in real time between two nodes.
And step 3, acquiring neighbor information features based on a neighbor transfer module to generate neighbor features of the node vector.
The neighbor transfer module acquires neighbor information features by adopting a transfer strategy along a time edge to generate neighbor features of the node vector.
In order to solve the problem of incomplete acquisition of high-order information, a neighbor delivery module is introduced. The neighbor transfer module acquires neighbor information based on a time edge transfer strategy, and then stores the neighbor information in a parallel operation mode; the neighbor transfer module essentially adopts a method based on a transfer strategy along a time edge, and is used for effectively acquiring the information characteristics of the neighbor nodes. This strategy helps capture high-level information to more fully characterize node features.
The neighbor delivery module independently stores a large amount of high-order neighbor information, and selectively extracts important neighbor information.
The specific process of the step3 is as follows:
And 3.1, transmitting the two-order neighbor information. The method comprises the following steps:
when the memory updating module generates a vector of real-time interaction between two nodes 、/>When the propagator firstly creates an interactive vector, the vector propagates to a vector list of k-hop neighbors through feedback operation, and the information of adjacent nodes is not required to be queried in the graph, and the calculation formula is as follows:
(7)；
(8)；
Wherein, 、/>Current moment/>, respectivelyUser node/>Commodity nodeNeighbor information of (2); /(I)The updating function is used for updating the information of the past nodes; /(I)、/>Current moment/>, respectivelyBy/>、/>The transferred node is embedded and generated by the multi-head attention module in the step 4.
When model reasoning is carried out, because the module independently stores a large amount of neighbor information, the method of the invention can be used for real-time model reasoning without inquiring the information of the adjacent nodes in the graph and only reading the stored information from the vector list of the adjacent nodes.
Step 3.2, carrying out vector list storage by adopting a neighbor generating function, and storing and transmitting neighbor information; the specific process is as follows:
(9)；
(10)；
Wherein, 、/>Current moment/>, respectivelyUser node/>Commodity node/>(1 /)Characteristic information of each neighbor node; /(I)Is the total number of neighbor nodes; /(I)To summarize the historical state functions of the neighbors of the user node or commodity node; /(I)、/>User node/>, respectively, at the last momentCommodity node/>Neighbor features of (a); /(I)Is an average strategy; /(I)As an attenuation or mapping function; /(I)The sampling strategy is passed for neighbor information.
The above-mentioned process is implemented by means of generating a functionGenerating neighbor information, and then adopting the neighbor information to transfer sampling strategy/>And transmitting the information to the neighbor nodes. Reasonably enhancing or attenuating the message and by averaging the strategy/>Multiple incoming neighbor information is aggregated into one to ensure balance.
And 3.3, performing position coding. In order to take the time arrival sequence of neighbors into consideration, position coding is performed, position information is converted into Onehot format, and then the position information is fed into the multi-head attention, and the specific process is as follows:
(11)；
(12)；
(13)；
(14)；
(15)；
(16)；
Wherein, 、/>Current moment/>, respectivelyUser node/>Commodity node/>Neighbor features after position coding; /(I)、/>Current moment/>, respectivelyUser node/>Commodity node/>Information vectors of different neighbor nodes;、/> Current moment/>, respectively User node/>Commodity node/>An absolute position encoding vector; /(I)、Current moment/>, respectivelyUser node/>Commodity node/>First/>Characteristic information of each neighbor node; /(I)、/>User nodes/>, respectivelyCommodity node/>First/>Absolute positions of the neighboring nodes.
The neighbor delivery module adopts a time stamp coding strategy to ensure that node messages are orderly transmitted to neighbor nodes according to the sequence of time stamps. This helps to better capture the timing characteristics of events in the dynamic diagram. The neighbor transfer module can ensure that efficient information transfer and updating can be realized in the dynamic graph.
And 4, generating node embedding based on the multi-head attention module.
The multi-head attention module creates current node embedding through multi-head attention operation on the basis of the memory updating module and the neighbor transmission module; the model utilizes a multi-headed attentiveness mechanism to enable it to learn knowledge of different aspects, thereby obtaining a richer representation of the node.
The forced model learns knowledge in different aspects, and multiple attention heads respectively calculate attention weights by using a plurality of independent attention heads, and splice or weight sum the results of the attention weights, so that richer representation is obtained;
User node Computing memory/>, corresponding to a message, at different timesAnd/>Input multi-head attention module generates user node embedding/>，/>For time/>User node/>Is a memory of (a); the multi-head attention module uses a multi-head attention mechanism to calculate the association between nodes, and calculates attention weights according to the matrix of three parameters of query vectors, key vectors and value vectors; the specific formula is as follows:
(17)；
(18)；
(19)；
(20)；
Node of commodity Computing memory/>, corresponding to a message, at different timesAnd/>Input multi-head attention module for generating commodity node embedding/>，/>For time/>Commodity node/>Is a memory of (a); the specific formula is as follows:
(21)；
(22)；
(23)；
(24)；
Wherein, As a multi-headed attention function; /(I)、/>User nodes/>, respectivelyCommodity node/>Is a query vector of (1); /(I)、/>User nodes/>, respectivelyCommodity node/>Is a key vector of (a); /(I)、/>User nodes/>, respectivelyCommodity node/>Is a vector of values of (a); /(I)Is an activation function; /(I)Transpose the symbol; /(I)Normalizing the data for a scaling factor, allowing the model to learn knowledge of different aspects; /(I)、/>User node/>, respectively, at the last momentCommodity node/>Is a memory of (a); /(I)、/>、/>、/>、/>、/>Respectively/>、/>、/>、/>、/>、/>Is used for capturing the interaction between the node information and the node neighbor information after the position coding.
And 5, decoding the obtained user node embedding and commodity node embedding by adopting a multi-layer perceptron, and obtaining and recommending commodities of interest to the user. Essentially, iterative updating of the node representation is achieved by a feedback strategy. The specific process is as follows:
step 5.1, calculating a predicted value of interest of the user to the commodity, wherein the specific formula is as follows:
(25)；
Wherein, A predicted value indicating whether or not the commodity is of interest is a predicted value of 0 to 1; /(I)Representing nonlinear activation functions, sigmoid activation functions, reLU activation functions, or the like can be adopted; /(I)、/>Are all learnable weights; /(I)Representing a splicing operation; /(I)、/>Are both learnable biases;
Step 5.2, presetting a threshold value, setting the threshold value to 0.5, if And if the threshold value is greater than or equal to the threshold value, the user node/>, is representedFor commodity node/>Interest, otherwise represent user node/>For commodity node/>Not of interest; by user node/>The prediction calculation is carried out with different commodity nodes, so that the user/>Is a plurality of items of interest.
The technical scheme of the invention can be used for learning the representation of the dynamic graph and is expected to be applied to various tasks of time sequence data analysis, dynamic network modeling and other related fields.
The memory updating module, the neighbor transferring module and the multi-head attention module form a neural network prediction model DIGE-MN for fusing memory updating and neighbor transferring. The model is trained by adopting a cross entropy loss function, and a time-varying negative sampling strategy is designed to construct positive and negative sample pairs. This is used to recommend items of interest to the user.
In order to demonstrate the feasibility and superiority of the invention, the following experiments were performed.
Experiment 1: and (5) comparing experiments. The present invention uses WIKIPEDIA, REDDIT, MOOC, ENRON, MEITUAN real-world dynamic graph datasets to evaluate the effectiveness of the DIGE-MN model of the present invention and compares it to JODIE, dyrep, evolveGCN, TGAT, TGN, APAN, MERIT seven dynamic data-aware models.
JODIE: two recurrent neural networks are used to update the embedding of users (users) and items (items) at each interaction. A novel projection operator is introduced that can learn and estimate the user's degree of embedding at any time in the future, and thus can simulate the future embedding trajectory of the user and the item.
Dyrep: taking into account the changes in network topology and node interactions, splitting a dynamic network into two processes: the association process and the communication process, the presentation learning can be regarded as an intermediary that effectively connects the two processes.
EvolveGCN: the graph rolling network model is employed in the time dimension without node embedding. The method evolves GCN parameters by using RNNs to capture the dynamics of the graph sequence.
TGAT: it is proposed that node embedding should include both static node features and varying topology features, and a novel functional time coding technique was developed based on self-attention mechanisms and according to classical Bochner theorem of harmonic analysis.
TGN: a generic, efficient dynamic graph model is presented in conjunction with memory modules and graph convolution operations, allowing the model to learn from the order of the data while maintaining efficient parallel processing.
APAN: an asynchronous propagation attention seeking neural network algorithm is proposed that decouples model reasoning from graph computation so that the graph computation is performed independently without affecting the speed of model reasoning.
MERIT: the most relevant neighbors are positioned in time sequence by adopting a context-aware attention mechanism, and a plurality of aggregations and propagation are finally executed by jointly capturing the contexts on the multi-level time diagram so as to explore and utilize the high-order structural information of the downstream task.
In the invention, the comparison experiment selects the AP and the AUC as evaluation indexes, the AP is the average precision, and the larger the AP value is, the better the performance is; AUC is the area covered by the ROC curve, with greater AUC values providing better performance.
Dividing seven dynamic data perception models into a discrete dynamic graph representation learning model and a continuous time dynamic graph representation learning model according to different processing modes of data; only EvolveGCN represents the learning model for the discrete dynamic diagram; JODIE, dyrep, TGAT, TGN, APAN, MERI is a continuous time dynamic graph representing a learning model. The evaluation results of each model are shown in table 1.
Table 1 comparison of the performance of the different models;
。
As can be seen from Table 1, the DIGE-MN model provided by the invention is always superior to all baseline models in recommending interesting commodities to users, compared with the models APAN and DIGE-MN which do not need to be memorized and updated, the DIGE-MN model provided by the invention improves the AP and AUC indexes by 6.63% and 4.01% at most, and the result shows that the node information has obvious improvement effect on the DIGE-MN through the construction of a memory updating unit.
Experiment 2: ablation experiments. In order to judge the contribution of each module used by the model to the final recommended performance of the model, the invention carries out ablative experiments on the model by deleting the test module. The present invention contemplates three variants of how much the DIGE-MN performance is improved for each pair of modules, with the symbolic use cases shown in Table 2.
Table 2 ablation experimental variation illustration;
。
The comparative results of the ablative experiments are shown in fig. 2 and 3. From FIGS. 2 and 3, it can be seen that the DIGE-MN model has performance advantages over four different data sets, particularly with respect to its variant model. This shows that different modules play different roles in improving the performance of the overall model. This also means that the effect of each module needs to be fully considered when building and optimizing the dynamic graph model to more fully mine the information in the dynamic graph.
Because the structure of the MOOC data set is special, fewer nodes are provided, but the interaction between the nodes is huge. In this case, the neighbor delivery module plays an important role in the performance of the data set, which underscores that the model may exhibit different advantages over different data sets due to the different data structures.
In summary, the results of ablation experiments highlight the excellent performance of the present invention over multiple datasets, while emphasizing the need for models to take into account dataset characteristics and interactions between different components in design and optimization.
In summary, the present invention has conducted comparative experiments on five real datasets, and the results indicate that DIGE-MN is superior to the existing most advanced methods, especially in recommending items of interest to users. In addition, the effectiveness of memory updating and neighbor delivery strategies is demonstrated through ablation experiments and parameter analysis. The method can realize the prediction recommendation of the dynamic perception data, has higher accuracy and stronger applicability, and provides technical support for the analysis and recommendation of the dynamic perception data.
In short, the invention is based on a dynamic graph neural network, and combines memory updating with innovative technology fusion developed by neighbor delivery neural network technology. The invention further focuses on the prediction application field after data processing, fuses the traditional machine learning method and the deep learning method and applies the fused traditional machine learning method and the deep learning method to different dynamic perception data sets, thereby realizing technical innovation and application innovation.
It should be understood that the above description is not intended to limit the invention to the particular embodiments disclosed, but to limit the invention to the particular embodiments disclosed, and that the invention is not limited to the particular embodiments disclosed, but is intended to cover modifications, adaptations, additions and alternatives falling within the spirit and scope of the invention.
Claims (6)
1. The method for recommending the e-commerce dynamic perception data based on memory updating and neighbor transfer is characterized by comprising the following steps of:
Step 1, acquiring commodity information, price, sales volume, evaluation and user behavior data from an electronic commerce platform, constructing an electronic commerce dynamic perception data set and preprocessing to obtain a time sequence data set;
step 2, inputting the time sequence data set into a memory updating module for capturing node information, aggregating information and updating memory;
Step 3, neighbor information characteristics are acquired based on a neighbor transfer module to generate neighbor characteristics of the node vector;
step 4, generating node embedding based on the multi-head attention module;
and 5, decoding the obtained node embedding by adopting a multi-layer perceptron to obtain the commodity which is recommended to be interested by the user and recommending the commodity.
2. The method for recommending e-commerce dynamic perception data based on memory updating and neighbor transfer according to claim 1, wherein the specific process of the step 1 is as follows:
step 1.1, acquiring commodity information, price, sales volume, evaluation and user behavior data through an API interface provided by an electronic commerce platform, and constructing and obtaining an electronic commerce dynamic perception data set from the data E-commerce dynamic perception dataset/>Is a graph network structure; wherein/>Representing e-commerce dynamic perception dataset/>User node in/>A user node sequence number; Representing e-commerce dynamic perception dataset/> Commodity node in/>The commodity node sequence number; /(I)Representing the current moment of the user commodity interaction event; /(I)Representing a duration of a user merchandise interaction event; /(I)Dynamic perception of datasets/>, for e-commerceMedium user node/>With commodity node/>Edges connected with each other; /(I)Representing the total number of user nodes,/>Representing the total number of commodity nodes;
Step 1.2, dynamically sensing data set for E-commerce Cleaning, and filling the missing values by using a linear interpolation method;
Step 1.3, extracting and sequencing data by utilizing numpy packets in python, storing edge characteristics, and finally obtaining a time sequence data set after digital signal processing ; Wherein the time series dataset/>Is a graph network structure; /(I)For time series dataset/>User nodes in (a); /(I)For time series dataset/>Commodity nodes in (a); /(I)Representing a time series dataset/>Medium user node/>With commodity node/>Edges connected with each other and corresponding to the current time/>User commodity interaction event.
3. The method for recommending e-commerce dynamic perception data based on memory updating and neighbor transfer according to claim 2, wherein the specific process of the step 2 is as follows:
Step 2.1, acquiring initial information of the node through a node information capturing unit, wherein a calculation formula is as follows:
(1)；
(2)；
Wherein, 、/>User nodes/>, respectivelyCommodity node/>At the present moment/>A calculated message; /(I)A multi-layer perceptron for a learnable message function; /(I)Representing the last moment of the commodity interaction event of the user; /(I)、/>User nodes/>, respectivelyCommodity node/>A memory vector of the previous moment;
Step 2.2, the aggregation of the user node information and the commodity node information is realized through an information aggregation unit, and the calculation formula is as follows:
(3)；
(4)；
Wherein, 、/>User nodes/>, respectivelyCommodity node/>An aggregated message; /(I) 、/>User nodes/>, respectivelyCommodity node/>At time/>A calculated message;
And 2.3, optimizing the storage of the user node and the commodity node through a memory updating unit, wherein the calculation formula is as follows:
(5)；
(6)；
Wherein, 、/>User nodes/>, respectivelyCommodity node/>Is a memory of (a); /(I)Is a learnable memory update function.
4. The method for recommending e-commerce dynamic perception data based on memory updating and neighbor delivery according to claim 3, wherein the specific process of the step 3 is as follows:
and 3.1, transmitting the two-order neighbor information, wherein the calculation formula is as follows:
(7)；
(8)；
Wherein, 、/>Current moment/>, respectivelyUser node/>Commodity node/>Neighbor information of (2); /(I)Is an update function; /(I)、/>Current moment/>, respectivelyBy/>、/>Embedding the transmitted nodes;
step 3.2, carrying out vector list storage by adopting a neighbor generating function, and storing and transmitting neighbor information; the specific process is as follows:
(9)；
(10)；
Wherein, 、/>Current moment/>, respectivelyUser node/>Commodity node/>First/>Characteristic information of each neighbor node; /(I)Is the total number of neighbor nodes; /(I)To summarize the historical state functions of the neighbors of the user node or commodity node;、/> User node/>, respectively, at the last moment Commodity node/>Neighbor features of (a); /(I)Is an average strategy; As an attenuation or mapping function; /(I) Transmitting a sampling strategy for neighbor information;
Step 3.3, performing position coding, wherein the specific process is as follows:
(11)；
(12)；
(13)；
(14)；
(15)；
(16)；
Wherein, 、/>Current moment/>, respectivelyUser node/>Commodity node/>Neighbor features after position coding; /(I)、/>Current moment/>, respectivelyUser node/>Commodity node/>Information vectors of different neighbor nodes;、/> Current moment/>, respectively User node/>Commodity node/>An absolute position encoding vector; /(I)、Current moment/>, respectivelyUser node/>Commodity node/>First/>Characteristic information of each neighbor node; /(I)、/>User nodes/>, respectivelyCommodity node/>First/>Absolute positions of the neighboring nodes.
5. The method for recommending e-commerce dynamic perception data based on memory updating and neighbor delivery according to claim 4, wherein the specific process of step 4 is as follows:
User node Computing memory/>, corresponding to a message, at different timesAnd (3) withInput multi-head attention module generates user node embedding/>，/>For time/>User node/>Is a memory of (a); the specific formula is as follows:
(17)；
(18)；
(19)；
(20)；
Node of commodity Computing memory/>, corresponding to a message, at different timesAnd (3) withInput multi-head attention module for generating commodity node embedding/>，/>For time/>Commodity node/>Is a memory of (a); the specific formula is as follows:
(21)；
(22)；
(23)；
(24)；
Wherein, As a multi-headed attention function; /(I)、/>User nodes/>, respectivelyCommodity node/>Is a query vector of (1); /(I)、/>User nodes/>, respectivelyCommodity node/>Is a key vector of (a); /(I)、/>User nodes/>, respectivelyCommodity node/>Is a vector of values of (a); /(I)Is an activation function; /(I)Transpose the symbol; /(I)Is a scaling factor; /(I)、/>User node/>, respectively, at the last momentCommodity node/>Is a memory of (a); /(I)、/>、/>、/>、/>、/>Respectively/>、/>、/>、/>、/>、/>Is a learning weight matrix.
6. The method for recommending e-commerce dynamic perception data based on memory updating and neighbor delivery according to claim 5, wherein the specific process of step 5 is as follows:
step 5.1, calculating a predicted value of interest of the user to the commodity, wherein the specific formula is as follows:
(25)；
Wherein, A predicted value indicating whether or not the commodity is of interest is a predicted value of 0 to 1; /(I)Representing a nonlinear activation function; /(I)、/>Are all learnable weights; /(I)Representing a splicing operation; /(I)、/>Are both learnable biases;
Step 5.2, presetting a threshold value if And if the threshold value is greater than or equal to the threshold value, the user node/>, is representedFor commodity node/>Interest, otherwise represent user node/>For commodity node/>Not of interest; by user node/>The prediction calculation is carried out with different commodity nodes, so that the user/>Is a plurality of items of interest.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410406224.6A CN117994011B (en) | 2024-04-07 | 2024-04-07 | E-commerce dynamic perception data recommendation method based on memory updating and neighbor transfer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410406224.6A CN117994011B (en) | 2024-04-07 | 2024-04-07 | E-commerce dynamic perception data recommendation method based on memory updating and neighbor transfer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117994011A true CN117994011A (en) | 2024-05-07 |
| CN117994011B CN117994011B (en) | 2024-07-05 |
Family
ID=90889256
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410406224.6A Active CN117994011B (en) | 2024-04-07 | 2024-04-07 | E-commerce dynamic perception data recommendation method based on memory updating and neighbor transfer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117994011B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118428993A (en) * | 2024-07-04 | 2024-08-02 | 青岛科技大学 | Personalized compatibility modeling method and system based on dynamic sampling and self-adaptive feature fusion |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104463637A (en) * | 2014-12-23 | 2015-03-25 | 北京石油化工学院 | Commodity recommendation method and device based on electronic business platform and server |
| CA3043686A1 (en) * | 2018-07-31 | 2020-01-31 | Middle Chart, LLC | Method and apparatus for augmented virtual models and orienteering |
| CN113919944A (en) * | 2021-09-07 | 2022-01-11 | 暨南大学 | Stock trading method and system based on reinforcement learning algorithm and time series model |
| US20220270155A1 (en) * | 2021-02-17 | 2022-08-25 | The Toronto-Dominion Bank | Recommendation with neighbor-aware hyperbolic embedding |
| CN115905721A (en) * | 2022-10-28 | 2023-04-04 | 天津大学 | Sequence recommendation algorithm based on time sequence perception graph attention network |
| CN116362836A (en) * | 2023-03-30 | 2023-06-30 | 东北林业大学 | Agricultural product recommendation algorithm based on user behavior sequence |
| CN116542742A (en) * | 2023-05-12 | 2023-08-04 | 广西师范大学 | Heterogeneous dynamic social recommendation method based on multiple relation types |
-
2024
- 2024-04-07 CN CN202410406224.6A patent/CN117994011B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104463637A (en) * | 2014-12-23 | 2015-03-25 | 北京石油化工学院 | Commodity recommendation method and device based on electronic business platform and server |
| CA3043686A1 (en) * | 2018-07-31 | 2020-01-31 | Middle Chart, LLC | Method and apparatus for augmented virtual models and orienteering |
| US20220270155A1 (en) * | 2021-02-17 | 2022-08-25 | The Toronto-Dominion Bank | Recommendation with neighbor-aware hyperbolic embedding |
| CN113919944A (en) * | 2021-09-07 | 2022-01-11 | 暨南大学 | Stock trading method and system based on reinforcement learning algorithm and time series model |
| CN115905721A (en) * | 2022-10-28 | 2023-04-04 | 天津大学 | Sequence recommendation algorithm based on time sequence perception graph attention network |
| CN116362836A (en) * | 2023-03-30 | 2023-06-30 | 东北林业大学 | Agricultural product recommendation algorithm based on user behavior sequence |
| CN116542742A (en) * | 2023-05-12 | 2023-08-04 | 广西师范大学 | Heterogeneous dynamic social recommendation method based on multiple relation types |
Non-Patent Citations (1)
| Title |
|---|
| 卢威: "基于图注意力网络的套餐推荐研究", 中国优秀硕士学位论文全文数据库, 30 November 2023 (2023-11-30), pages 36 - 46 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118428993A (en) * | 2024-07-04 | 2024-08-02 | 青岛科技大学 | Personalized compatibility modeling method and system based on dynamic sampling and self-adaptive feature fusion |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117994011B (en) | 2024-07-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pan et al. | Study on convolutional neural network and its application in data mining and sales forecasting for E-commerce | |
| CN112364976B (en) | User preference prediction method based on session recommendation system | |
| CN111797321B (en) | Personalized knowledge recommendation method and system for different scenes | |
| CN117994011B (en) | E-commerce dynamic perception data recommendation method based on memory updating and neighbor transfer | |
| CN108563755A (en) | A kind of personalized recommendation system and method based on bidirectional circulating neural network | |
| Liu et al. | GNNRec: Gated graph neural network for session-based social recommendation model | |
| CN114265986B (en) | Information pushing method and system fusing knowledge graph structure and path semantics | |
| CN114971784B (en) | Session recommendation method and system based on graph neural network by fusing self-attention mechanism | |
| CN114637863B (en) | Knowledge graph recommendation method based on propagation | |
| CN113590976A (en) | Recommendation method of space self-adaptive graph convolution network | |
| CN115860880B (en) | Personalized commodity recommendation method and system based on multi-layer heterogeneous graph convolution model | |
| CN115017405B (en) | Graph neural network travel package recommendation method based on multi-task self-coding | |
| CN116304279B (en) | Active perception method and system for evolution of user preference based on graph neural network | |
| CN115470406A (en) | Graph neural network session recommendation method based on dual-channel information fusion | |
| CN114925270B (en) | Session recommendation method and model | |
| CN114282077A (en) | Session recommendation method and system based on session data | |
| CN115481325A (en) | Personalized news recommendation method and system based on user global interest migration perception | |
| CN115631008A (en) | Commodity recommendation method, commodity recommendation device, commodity recommendation equipment and commodity recommendation medium | |
| CN116166875A (en) | Bidirectional cross-domain recommendation method of heterogeneous graph neural network based on element path enhancement | |
| CN113240086A (en) | Complex network link prediction method and system | |
| CN116308854A (en) | Information cascading popularity prediction method and system based on probability diffusion | |
| Sang et al. | Position-aware graph neural network for session-based recommendation | |
| CN114491267A (en) | Article recommendation method and device and storage medium | |
| CN117194765A (en) | Dual-channel graph contrast learning session recommendation method for interest perception | |
| CN116975686A (en) | Method for training student model, behavior prediction method and device |
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 |