CN114169975A - Shopping network commodity recommendation method and system based on random walk heterogeneous attention - Google Patents
Shopping network commodity recommendation method and system based on random walk heterogeneous attention Download PDFInfo
- Publication number
- CN114169975A CN114169975A CN202111551744.9A CN202111551744A CN114169975A CN 114169975 A CN114169975 A CN 114169975A CN 202111551744 A CN202111551744 A CN 202111551744A CN 114169975 A CN114169975 A CN 114169975A
- Authority
- CN
- China
- Prior art keywords
- node
- commodity
- attention
- user
- heterogeneous
- 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
- 238000005295 random walk Methods 0.000 title claims abstract description 36
- 238000000034 method Methods 0.000 title claims abstract description 32
- 239000013604 expression vector Substances 0.000 claims abstract description 41
- 238000004364 calculation method Methods 0.000 claims abstract description 30
- 238000005457 optimization Methods 0.000 claims abstract description 19
- 238000010276 construction Methods 0.000 claims abstract description 12
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 9
- 239000013598 vector Substances 0.000 claims description 67
- 230000005012 migration Effects 0.000 claims description 24
- 238000013508 migration Methods 0.000 claims description 24
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 9
- 230000002776 aggregation Effects 0.000 claims description 8
- 238000004220 aggregation Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 230000007246 mechanism Effects 0.000 claims description 6
- 230000004931 aggregating effect Effects 0.000 claims description 5
- 238000011478 gradient descent method Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 3
- XHEFDIBZLJXQHF-UHFFFAOYSA-N fisetin Chemical compound C=1C(O)=CC=C(C(C=2O)=O)C=1OC=2C1=CC=C(O)C(O)=C1 XHEFDIBZLJXQHF-UHFFFAOYSA-N 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000002347 injection Methods 0.000 description 1
- 239000007924 injection Substances 0.000 description 1
- 239000010985 leather Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
Images
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
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06N3/048—Activation functions
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Mathematical Physics (AREA)
- Finance (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (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 relates to a shopping network commodity recommendation system based on random walk heterogeneous attention, which comprises a heterogeneous network construction module, a random walk module, an intra-class attention coefficient calculation module, an inter-class attention coefficient calculation module, a node expression vector optimization module and a clustering and commodity recommendation module, wherein the random walk module is used for calculating intra-class attention coefficients of different classes; the shopping network is modeled into a heterogeneous network comprising user nodes and commodity nodes, heterogeneous information is fused by using random walk and heterogeneous attention, and finally communities formed by related commodities are obtained by combining a clustering algorithm, so that the method is used for personalized recommendation to improve commodity recommendation rationality.
Description
Technical Field
The invention relates to the field of big data analysis, in particular to a shopping network commodity recommendation method and system based on random walk heterogeneous attention.
Background
With the development of the times and the advancement of technology, especially the information technology in the third industrial leather, many daily lives of people are shifted to online, and online shopping is a typical example. The online shopping is connected with consumers and producers through the Internet, so that the consumers can purchase psychographic products without going out of households, and the producers can sell the products to the consumers all over the country. The convenience of online shopping makes the online shopping continuously developed, and in the twenty-one of 2021, the final bargain amount of the online mart of the Tianmao reaches 5403 billion yuan. The rapid development also makes the shopping network very complicated, and how to carry out accurate commodity recommendation on users has important significance for consumers and producers. However, the traditional shopping network commodity recommendation method does not perform refined modeling on the shopping network, mainly recommends similar commodities of the same kind according to the purchased commodities, and is worthless for the user who has purchased the commodities.
Disclosure of Invention
In view of this, the invention aims to provide a shopping network commodity recommendation method and system based on random walk heterogeneous attention, which perform more refined modeling on a shopping network, construct a heterogeneous network including user nodes and commodity nodes, and effectively improve the commodity recommendation accuracy by considering heterogeneous information.
In order to achieve the purpose, the invention adopts the following technical scheme:
a shopping network commodity recommendation method based on random walk heterogeneous attention comprises the following steps:
step S1, constructing a heterogeneous network containing user nodes and commodity nodes according to the shopping records of the users;
step S2, traversing all nodes on the heterogeneous network based on the heterogeneous network, taking each node as a starting point of the migration, performing random migration for a plurality of times, recording the access times of each node in the migration process, arranging the access times in a descending order, and selecting the front node as a closely related neighbor node set of the starting point according to a preset proportion;
step S3: for each node, respectively executing in a user node neighbor set commodity node neighbor set to obtain a user expression vector and a commodity expression vector;
step S4, calculating attention coefficients of user representation vectors and commodity representation vectors for each node, representing different importance of user node neighbors and commodity node neighbors to the migration starting point, and finally performing weighted aggregation to generate final representation vectors of the migration starting point;
step S5, calculating a loss function and performing backward propagation according to the final expression vector, and optimizing by using a random gradient descent method to obtain an optimized node expression vector;
step S6: according to the optimized node expression vector, the expression vector of the commodity node is processed by using a K-means algorithm to obtain a commodity set with closely related inner parts, an interested commodity set is determined from commodities in a shopping record of a user who wants to recommend the commodities, and the commodities in the commodity set are recommended to the user.
Further, the step S3 is specifically:
step S31: classifying the closely related neighbor node sets obtained in the step S2 according to the types of the nodes, dividing the closely related neighbor node sets into a user node set and a commodity node set, and calculating the intra-class attention coefficient in each type;
step S32: in the same node type, the information of different nodes is aggregated by using attention, which is called an in-class attention mechanism; in this way, a representation vector representing the influence of the neighbor of the type is generated;
step S33: on the basis of the attention coefficient calculated at step S32, a representative vector representing such neighbors is calculated using the following formula.
Where σ denotes a non-linear sigmoid function and W is a weight matrix.
Further, the step S32 isThe body is as follows: if the type of the currently processed node is t, the node v in the typeiAnd node vjThe attention coefficient therebetween is calculated as follows:
where | | | denotes join operation, W is a weight matrix, LeakReLU is an activation function, and vector α is a parameter that can be learned in training, hiIs node viThe hidden layer vector of (1).
Further, the step S33 uses a multi-head attention mechanism, and the final output is an average of multiple attention head output vectors or is directly spliced.
Further, the step S4 is specifically:
step S41: aggregating the user node expression vector and the commodity node expression vector, wherein the specific calculation formula is as follows:
wherein the vector γ is one of the parameters learned in training; w is the weight matrix used for the linear transformation, and b is the offset vector that plays a role in the offset; attention coefficient a of node type ttNormalization is performed using a softmax function;
step S42: node viThe final representation vector is represented by each node type representation vector zi tBased on inter-class attention coefficient atAnd generating weighted aggregation.
Further, the step S5 specifically includes the following steps:
for unsupervised associated commodity set discovery tasks, the following loss function is used as the objective function for training:
wherein
Is shown at the sum node viSet of closely related nodes NiAnd the node type is t. Theta represents the parameters of the model, and the definition of p (j | i; theta) is shown in the following formula
Using a negative sampling technique to speed up, the final loss function is as follows:
where node k is a set of nodes that are not present
Negative sample node in (1).
Further, the step S5 specifically includes the following steps:
and (3) processing the expression vectors of the commodity nodes by using a K-means algorithm, and minimizing a cost function by using an iterative optimization method to obtain a commodity set with closely related internal parts, wherein the optimization function is as follows:
wherein u isjRepresenting the vector, x, for the center in the set of items CiThe expression vector of each commodity in the commodity set is shown, and n is the number of commodities in the commodity set.
After each commodity set is obtained through iterative optimization, an interested commodity set is determined from commodities in a shopping record of a user who wants to recommend the commodities, and the commodities in the commodity set are recommended to the user.
A shopping network commodity recommendation system based on random walk heterogeneous attention comprises a heterogeneous network construction module, a random walk module, an intra-class attention coefficient calculation module, an inter-class attention coefficient calculation module, a node expression vector optimization module and a clustering and commodity recommendation module,
the heterogeneous network construction module starts from a user who wants to recommend commodities, constructs a heterogeneous network comprising user nodes and commodity nodes according to shopping records, and executes the heterogeneous network construction module once every preset time;
the random walk module executes random walk on the heterogeneous network generated by the heterogeneous network construction module;
the intra-class attention coefficient calculation module is used for respectively executing neighbor centralized execution on the commodity node neighbor set of the user node neighbor set for each node on the basis of the output of the random walk module; in the user node neighbor set, calculating the attention coefficient of each node by using an attention layer, finally weighting and aggregating the vectors of each node, generating a user representation vector representing the user node neighbor, and representing the influence of the neighbor node of the user node type on the wandering starting point; similarly, commodity expression vectors are independently generated in the commodity node set to express the influence of neighbor nodes of the commodity node type on the wandering starting point; the intra-class attention coefficient calculation module outputs two vectors which are a user representation vector and a commodity representation vector respectively;
the inter-class attention coefficient calculation module calculates the attention coefficients of user representation vectors and commodity representation vectors by using an attention layer for each node on the basis of the output of the intra-class attention calculation module, represents different importance of user node neighbors and commodity node neighbors to the migration starting point, and finally performs weighted aggregation to generate a final representation vector of the migration starting point;
the node expression vector optimization module calculates a loss function and performs back propagation by using a final expression vector output by the inter-class attention coefficient calculation module, optimizes parameters in the inter-class attention coefficient calculation module and the intra-class attention coefficient calculation module by using a random gradient descent method, designs a loss function based on positive and negative samples, and performs iterative optimization in an unsupervised mode;
the clustering and commodity recommending module processes the expression vectors of the commodity nodes by using a K-means algorithm on the basis of the node expression vectors optimized by the node expression vector optimizing module to obtain a commodity set with closely related inner parts, determines an interested commodity set from commodities in a shopping record of a user who needs commodity recommendation, and recommends the commodities in the commodity set to the user.
Further, the random walk module specifically includes: traversing all nodes on the heterogeneous network, taking each node as a starting point of the migration, performing multiple random migrations, recording the access times of each node in the migration process, arranging the access times in a descending order, and selecting the previous node as a closely related neighbor node set of the starting point; the selected proportion is controlled by a hyper-parameter, and a proper balance is searched in insufficient information and introduced noise; and for the close neighbor node set of each node, separating the nodes according to the user nodes and the commodity nodes.
Compared with the prior art, the invention has the following beneficial effects:
1. according to the invention, a shopping network is more finely modeled, a heterogeneous network comprising user nodes and commodity nodes is constructed, and the accuracy of commodity recommendation is effectively improved by considering heterogeneous information;
2. the invention does not need prior knowledge to preset the meta path, and is more convenient in practical application.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
Referring to fig. 1, the present invention provides a shopping network commodity recommendation method based on random walk heterogeneous attention, and provides a system, including: the system comprises a heterogeneous network construction module, a random walk module, an intra-class attention coefficient calculation module, an inter-class attention coefficient calculation module, a node expression vector optimization module and a clustering and commodity recommendation module;
step S1: the heterogeneous network construction module constructs a heterogeneous network comprising user nodes and commodity nodes according to shopping records from users who want to recommend commodities. Since the user may continuously purchase new commodities, the heterogeneous network construction module needs to be executed at intervals, so that the accuracy of subsequent commodity recommendation is improved;
step S2: the random walk module performs random walks on the heterogeneous network generated at step S1. Specifically, all nodes on the heterogeneous network are traversed, and each node is taken as a starting point of the migration, and multiple random migrations are performed. And recording the access times of each node in the walking process, arranging the nodes in a descending order, and selecting the previous node as a closely related neighbor node set of a starting point. The selected ratio is controlled by a hyper-parameter to find a suitable balance between insufficient information and introduced noise. And for the close neighbor node set of each node, the nodes are separated according to the user node and the commodity node, so that the processing of subsequent steps is facilitated.
Step S3: the intra-class attention coefficient calculation module is respectively executed in the user node neighbor set commodity node neighbor set for each node on the basis of the output of the random walk module in the step S2. In the user node neighbor set, the attention layer is used for calculating the attention coefficient of each node, finally, the vectors of each node are weighted and aggregated, and a user representation vector representing the user node neighbor is generated and represents the influence of the neighbor node of the user node type on the wandering starting point. Similarly, commodity expression vectors are independently generated in the commodity node set and represent the influence of the neighbor nodes of the commodity node types on the wandering starting point. The intra-class attention coefficient calculation module outputs two vectors, namely a user representation vector and a commodity representation vector.
Step S4: the inter-class attention coefficient calculating module calculates, for each node, the attention coefficients of the user representation vector and the commodity representation vector using the attention layer on the basis of the output of the intra-class attention calculating module in step S3, and represents different importance of the user node neighbor and the commodity node neighbor to the migration start point, and finally performs weighted aggregation to generate a final representation vector of the migration start point. By means of the method of classifying and processing according to the node types, heterogeneous information that nodes in a heterogeneous network have different types is considered, so that node expression vectors are generated more finely, and the quality of the node expression vectors is improved.
Step S5: the nodes represent a vector optimization module, the nodes represent vectors generated in the step S4, a loss function is calculated and propagated in reverse, and parameters in the inter-class attention coefficient calculation module in the step S3 and the intra-class attention coefficient calculation module in the step S4 are optimized by using a stochastic gradient descent method. By designing a loss function based on positive and negative samples, iterative optimization can be carried out in an unsupervised mode, and particularly, the node representation vector is similar to the nodes of the positive samples and is not similar to the nodes of the negative samples.
Step S6: the clustering and commodity recommending module processes the expression vectors of the commodity nodes by using a K-means algorithm on the basis of the optimized node expression vectors in the step S5 to obtain a commodity set with closely related inner parts. And aiming at the user who needs to recommend the commodities, determining an interested commodity set from the commodities in the shopping record of the user, and recommending the commodities in the commodity set to the user.
Preferably, in this embodiment, the step S2 specifically includes the following steps:
step S21: each node on the heterogeneous network can be used as a starting point of random walk, and a neighbor node set closely related to the node is searched through multiple random walks. Meanwhile, the accessed nodes and the access times are recorded in the walking process and are arranged in a descending order according to the accessed frequency.
Step S22: from the result obtained in step S21, the node with the highest access frequency is selected as the input of the subsequent processing. The specific selection ratio is controlled by the hyperparameter β. For example, when the value of the parameter β is set to 0.1, only the top 10% of the nodes obtained in step S21 are selected as the input of the subsequent processing.
Preferably, in this embodiment, the step S3 specifically includes the following steps:
step S31: and classifying the closely related node sets output in the step S22 according to the types of the nodes, dividing the node sets into a user node set and a commodity node set, and calculating the intra-class attention coefficient in each type.
Step S32: within the same node type, we use attention to aggregate information of different nodes, which is called an intra-class attention mechanism. In this way, a representation vector representing this type of neighbor influence is generated. Assuming that the type of the currently processed node is t, the node v in the typeiAnd node vjThe attention coefficient therebetween is calculated as follows:
where | | represents join operations, W is a weight matrix, LeakReLU is an activation function, and vector α is a parameter that may be learned in training. h isiIs node viThe hidden layer vector of (1).
Step S33: on the basis of the attention coefficient calculated at step S32, a representative vector representing such neighbors is calculated using the following formula.
Where σ denotes a non-linear sigmoid function and W is a weight matrix. In addition, in order to make the result more stable, a multi-head attention mechanism is used, and the final output can be the average of a plurality of attention head output vectors or directly spliced.
Preferably, in this embodiment, the step S4 specifically includes the following steps:
step S41: and aggregating the user node representation vectors and the commodity node representation vectors output in the step S33 by using another attention layer, wherein a specific calculation formula is as follows:
where the vector gamma is one of the parameters learned in the training. W is the weight matrix used for the linear transformation and b is the offset vector that acts as the offset. Attention coefficient a of node type ttLike the injection force coefficient in the formula (4-1)
Also, the softmax function is used for normalization.
Step S42: node viThe final representation vector is represented by each node type representation vector zi tBased on inter-class attention coefficient atAnd generating weighted aggregation.
Preferably, in this embodiment, the step S5 specifically includes the following steps:
step S51: for unsupervised associated commodity set discovery tasks, the following loss function is used as the objective function for training:
wherein
Is shown at the sum node viSet of closely related nodes NiAnd the node type is t. Theta represents the parameters of the model, and the definition of p (j | i; theta) is shown in the following formula.
Step S52: in practice, the calculation of the formula in step S51 is time consuming, and uses a negative sampling technique to speed up, and the final loss function is as follows:
where node k is a set of nodes that are not present
Negative sample node in (1).
Step S6: and (3) processing the expression vectors of the commodity nodes by using a K-means algorithm, and minimizing a cost function by using an iterative optimization method to obtain a commodity set with closely related internal parts, wherein the optimization function is as follows:
wherein u isjRepresenting the vector, x, for the center in the set of items CiRepresenting vectors of all commodities in the commodity set, wherein n is the number of the commodities in the commodity set;
after each commodity set is obtained through iterative optimization, an interested commodity set is determined from commodities in a shopping record of a user who wants to recommend the commodities, and the commodities in the commodity set are recommended to the user.
The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.
Claims (9)
1. A shopping network commodity recommendation method based on random walk heterogeneous attention is characterized by comprising the following steps:
step S1, constructing a heterogeneous network containing user nodes and commodity nodes according to the shopping records of the users;
step S2, traversing all nodes on the heterogeneous network based on the heterogeneous network, taking each node as a starting point of the migration, performing random migration for a plurality of times, recording the access times of each node in the migration process, arranging the access times in a descending order, and selecting the front node as a closely related neighbor node set of the starting point according to a preset proportion;
step S3: for each node, respectively executing in a user node neighbor set commodity node neighbor set to obtain a user expression vector and a commodity expression vector;
step S4, calculating attention coefficients of user representation vectors and commodity representation vectors for each node, representing different importance of user node neighbors and commodity node neighbors to the migration starting point, and finally performing weighted aggregation to generate final representation vectors of the migration starting point;
step S5, calculating a loss function and performing backward propagation according to the final expression vector, and optimizing by using a random gradient descent method to obtain an optimized node expression vector;
step S6: according to the optimized node expression vector, the expression vector of the commodity node is processed by using a K-means algorithm to obtain a commodity set with closely related inner parts, an interested commodity set is determined from commodities in a shopping record of a user who wants to recommend the commodities, and the commodities in the commodity set are recommended to the user.
2. The shopping network commodity recommendation method based on random walk heterogeneous attention of claim 1, wherein the step S3 is specifically:
step S31: classifying the closely related neighbor node sets obtained in the step S2 according to the types of the nodes, dividing the closely related neighbor node sets into a user node set and a commodity node set, and calculating the intra-class attention coefficient in each type;
step S32: in the same node type, the information of different nodes is aggregated by using attention, which is called an in-class attention mechanism; in this way, a representation vector representing the influence of the neighbor of the type is generated;
step S33: on the basis of the attention coefficient calculated at step S32, a representative vector representing such neighbors is calculated using the following formula.
Where σ denotes a non-linear sigmoid function and W is a weight matrix.
3. The shopping network commodity recommendation method based on random walk heterogeneous attention of claim 1, wherein the step S32 is specifically: if the type of the currently processed node is t, the node v in the typeiAnd node vjThe attention coefficient therebetween is calculated as follows:
where | | | denotes join operation, W is a weight matrix, LeakReLU is an activation function, and vector α is a parameter that can be learned in training, hiIs node viThe hidden layer vector of (1).
4. The shopping network commodity recommendation method based on random walk heterogeneous attention of claim 1, wherein the step S33 uses a multi-head attention mechanism, and the final output is an average of a plurality of attention head output vectors or is directly spliced.
5. The shopping network commodity recommendation method based on random walk heterogeneous attention of claim 1, wherein the step S4 is specifically:
step S41: aggregating the user node expression vector and the commodity node expression vector, wherein the specific calculation formula is as follows:
wherein the vector γ is one of the parameters learned in training; w is the weight matrix used for the linear transformation, and b is the offset vector that plays a role in the offset; attention of node type tCoefficient atNormalization is performed using a softmax function;
step S42: node viThe final representation vector is represented by each node type representation vector zi tBased on inter-class attention coefficient atAnd generating weighted aggregation.
6. The shopping network commodity recommendation method based on random walk heterogeneous attention according to claim 1, wherein the step S5 specifically comprises the steps of:
for unsupervised associated commodity set discovery tasks, the following loss function is used as the objective function for training:
wherein
Is shown at the sum node viSet of closely related nodes NiAnd the node type is t. Theta represents the parameters of the model, and the definition of p (j | i; theta) is shown in the following formula
Using a negative sampling technique to speed up, the final loss function is as follows:
where node k is a set of nodes that are not present
Negative sample node in (1).
7. The shopping network commodity recommendation method based on random walk heterogeneous attention according to claim 1, wherein the step S6 specifically comprises the steps of:
and (3) processing the expression vectors of the commodity nodes by using a K-means algorithm, and minimizing a cost function by using an iterative optimization method to obtain a commodity set with closely related internal parts, wherein the optimization function is as follows:
wherein u isjRepresenting the vector, x, for the center in the set of items CiRepresenting vectors of all commodities in the commodity set, wherein n is the number of the commodities in the commodity set;
after each commodity set is obtained through iterative optimization, an interested commodity set is determined from commodities in a shopping record of a user who wants to recommend the commodities, and the commodities in the commodity set are recommended to the user.
8. A shopping network commodity recommendation system based on random walk heterogeneous attention is characterized by comprising a heterogeneous network construction module, a random walk module, an intra-class attention coefficient calculation module, an inter-class attention coefficient calculation module, a node expression vector optimization module and a clustering and commodity recommendation module,
the heterogeneous network construction module starts from a user who wants to recommend commodities, constructs a heterogeneous network comprising user nodes and commodity nodes according to shopping records, and executes the heterogeneous network construction module once every preset time;
the random walk module executes random walk on the heterogeneous network generated by the heterogeneous network construction module;
the intra-class attention coefficient calculation module is used for respectively executing neighbor centralized execution on the commodity node neighbor set of the user node neighbor set for each node on the basis of the output of the random walk module; in the user node neighbor set, calculating the attention coefficient of each node by using an attention layer, finally weighting and aggregating the vectors of each node, generating a user representation vector representing the user node neighbor, and representing the influence of the neighbor node of the user node type on the wandering starting point; similarly, commodity expression vectors are independently generated in the commodity node set to express the influence of neighbor nodes of the commodity node type on the wandering starting point; the intra-class attention coefficient calculation module outputs two vectors which are a user representation vector and a commodity representation vector respectively;
the inter-class attention coefficient calculation module calculates the attention coefficients of user representation vectors and commodity representation vectors by using an attention layer for each node on the basis of the output of the intra-class attention calculation module, represents different importance of user node neighbors and commodity node neighbors to the migration starting point, and finally performs weighted aggregation to generate a final representation vector of the migration starting point;
the node expression vector optimization module calculates a loss function and performs back propagation by using a final expression vector output by the inter-class attention coefficient calculation module, optimizes parameters in the inter-class attention coefficient calculation module and the intra-class attention coefficient calculation module by using a random gradient descent method, designs a loss function based on positive and negative samples, and performs iterative optimization in an unsupervised mode;
the clustering and commodity recommending module processes the expression vectors of the commodity nodes by using a K-means algorithm, for example, on the basis of the node expression vectors optimized by the node expression vector optimizing module to obtain a commodity set with closely related inner parts, determines an interested commodity set from commodities in a shopping record of a user who needs commodity recommendation, and recommends the commodities in the commodity set to the user.
9. The shopping network commodity recommendation system based on random walk heterogeneous attention of claim 8, wherein the random walk module is specifically: traversing all nodes on the heterogeneous network, taking each node as a starting point of the migration, performing multiple random migrations, recording the access times of each node in the migration process, arranging the access times in a descending order, and selecting the previous node as a closely related neighbor node set of the starting point; the selected proportion is controlled by a hyper-parameter, and a proper balance is searched in insufficient information and introduced noise; and for the close neighbor node set of each node, separating the nodes according to the user nodes and the commodity nodes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111551744.9A CN114169975B (en) | 2021-12-17 | 2021-12-17 | Shopping network commodity recommendation method and system based on random walk heterogeneous attention |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111551744.9A CN114169975B (en) | 2021-12-17 | 2021-12-17 | Shopping network commodity recommendation method and system based on random walk heterogeneous attention |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114169975A true CN114169975A (en) | 2022-03-11 |
| CN114169975B CN114169975B (en) | 2024-07-19 |
Family
ID=80487271
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111551744.9A Active CN114169975B (en) | 2021-12-17 | 2021-12-17 | Shopping network commodity recommendation method and system based on random walk heterogeneous attention |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114169975B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114707066A (en) * | 2022-04-01 | 2022-07-05 | 福州大学 | Scenic spot recommendation method and system based on community perception and self-adaptive random walk |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108320218A (en) * | 2018-02-05 | 2018-07-24 | 湖南大学 | Individual commodity recommendation method based on trust-scoring time-evolution two-way effect |
| CN108898459A (en) * | 2018-06-25 | 2018-11-27 | 中国联合网络通信集团有限公司 | A kind of Method of Commodity Recommendation and device |
| CN111177781A (en) * | 2019-12-30 | 2020-05-19 | 北京航空航天大学 | Differential privacy recommendation method based on heterogeneous information network embedding |
| CN112069415A (en) * | 2020-08-13 | 2020-12-11 | 中国海洋大学 | Interest point recommendation method based on heterogeneous attribute network characterization learning |
| WO2021103789A1 (en) * | 2019-11-27 | 2021-06-03 | 西北工业大学 | Personalized accurate recommendation method driven by knowledge graph |
-
2021
- 2021-12-17 CN CN202111551744.9A patent/CN114169975B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108320218A (en) * | 2018-02-05 | 2018-07-24 | 湖南大学 | Individual commodity recommendation method based on trust-scoring time-evolution two-way effect |
| CN108898459A (en) * | 2018-06-25 | 2018-11-27 | 中国联合网络通信集团有限公司 | A kind of Method of Commodity Recommendation and device |
| WO2021103789A1 (en) * | 2019-11-27 | 2021-06-03 | 西北工业大学 | Personalized accurate recommendation method driven by knowledge graph |
| CN111177781A (en) * | 2019-12-30 | 2020-05-19 | 北京航空航天大学 | Differential privacy recommendation method based on heterogeneous information network embedding |
| CN112069415A (en) * | 2020-08-13 | 2020-12-11 | 中国海洋大学 | Interest point recommendation method based on heterogeneous attribute network characterization learning |
Non-Patent Citations (1)
| Title |
|---|
| 王宏琳: "基于随机游走异构注意力的购物网络商品推荐方法及系统", 计算机研究与发展, vol. 59, no. 07, 26 August 2021 (2021-08-26), pages 1510 - 1520 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114707066A (en) * | 2022-04-01 | 2022-07-05 | 福州大学 | Scenic spot recommendation method and system based on community perception and self-adaptive random walk |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114169975B (en) | 2024-07-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108648049B (en) | Sequence recommendation method based on user behavior difference modeling | |
| Patro et al. | A hybrid action-related K-nearest neighbour (HAR-KNN) approach for recommendation systems | |
| CN110910218B (en) | Multi-behavior migration recommendation method based on deep learning | |
| CN111062775A (en) | Recommendation system recall method based on attention mechanism | |
| CN107220365A (en) | Accurate commending system and method based on collaborative filtering and correlation rule parallel processing | |
| Yu et al. | Cross-domain recommendation based on latent factor alignment | |
| CN115186097A (en) | Knowledge graph and reinforcement learning based interactive recommendation method | |
| CN111199458A (en) | Recommendation system based on meta-learning and reinforcement learning | |
| CN108563690A (en) | A kind of collaborative filtering recommending method based on object-oriented cluster | |
| CN114329200B (en) | Personalized self-adaptive binding recommendation model and recommendation method | |
| Ambulgekar et al. | A survey on collaborative filtering: tasks, approaches and applications | |
| CN112231583A (en) | E-commerce recommendation method based on dynamic interest group identification and generation of countermeasure network | |
| CN114169975A (en) | Shopping network commodity recommendation method and system based on random walk heterogeneous attention | |
| Pang et al. | Efficient deep reinforcement learning-enabled recommendation | |
| Wu et al. | Result Diversification in Search and Recommendation: A Survey | |
| Yu et al. | A graph attention network under probabilistic linguistic environment based on Bi-LSTM applied to film classification | |
| Gasmi et al. | Context-aware based evolutionary collaborative filtering algorithm | |
| CN116069921A (en) | News recommendation method integrating activation diffusion theory and Eggy-Cursor theory | |
| CN115687757A (en) | Recommendation method fusing hierarchical attention and feature interaction and application system thereof | |
| CN115344794A (en) | Scenic spot recommendation method based on knowledge map semantic embedding | |
| CN115048530A (en) | Graph convolution recommendation system fusing neighbor importance and feature learning | |
| Yi et al. | DMMP: A distillation-based multi-task multi-tower learning model for personalized recommendation | |
| Qiu et al. | Multi-view hybrid recommendation model based on deep learning | |
| CN110956528A (en) | Recommendation method and system for e-commerce platform | |
| Liu | Product image recommendation with transformer model using deep reinforcement learning |
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 |