CN112861006A - Recommendation method and system fusing meta-path semantics - Google Patents

Recommendation method and system fusing meta-path semantics Download PDF

Info

Publication number
CN112861006A
CN112861006A CN202110198542.4A CN202110198542A CN112861006A CN 112861006 A CN112861006 A CN 112861006A CN 202110198542 A CN202110198542 A CN 202110198542A CN 112861006 A CN112861006 A CN 112861006A
Authority
CN
China
Prior art keywords
user
commodity
node
meta
nodes
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
Application number
CN202110198542.4A
Other languages
Chinese (zh)
Other versions
CN112861006B (en
Inventor
许倩倩
郝前秀
杨智勇
黄庆明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Institute of Computing Technology of CAS
Original Assignee
Institute of Computing Technology of CAS
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Institute of Computing Technology of CAS filed Critical Institute of Computing Technology of CAS
Priority to CN202110198542.4A priority Critical patent/CN112861006B/en
Publication of CN112861006A publication Critical patent/CN112861006A/en
Application granted granted Critical
Publication of CN112861006B publication Critical patent/CN112861006B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention provides a recommendation method fusing meta-path semantics, which comprises the following steps: calculating a relative importance bipartite graph of the user/commodity based on a heterogeneous information network and a user/commodity scoring data set; guiding information propagation between user/commodity nodes by using the relative importance bipartite graph to obtain embedded expression of the user/commodity nodes; and transforming the embedded expression of the user/commodity node to obtain a final embedded expression of the user/commodity node for recommendation, and inputting the final embedded expression of the user/commodity node into a recommendation model to obtain the prediction score of the user/commodity.

Description

Recommendation method and system fusing meta-path semantics
Technical Field
The invention relates to the technical field of computer networks, in particular to a recommendation method and a recommendation system fusing meta-path semantics.
Background
With the rapid development of the internet, internet users face massive information every day, and under the condition that the users do not have specific purposes, commodities which are possibly interested by the users are selected from the massive commodities to be recommended, so that the potential interests of the users can be greatly stimulated, the users can be helped to make decisions efficiently under the condition of information overload, and the method becomes an important growth engine for the development of the internet. Accordingly, recommendation systems play an increasingly important role in various internet products.
Although there are many algorithms that have been successfully applied to recommendation systems, most of them focus on learning recommendation strategies from a single user-commodity interaction record (or plus simple content information). This will inevitably face problems with data sparseness, cold starts, over recommendations, etc. In recent years, some researches indicate that other relationship data between users and other relationship data between the users and other relationship data between commodities can help a recommendation system to solve one or more of the problems; some researches introduce social networks into recommendation systems to solve problems caused by sparse user transaction record data; to further exploit the more complex, heterogeneous user/commodity relationship data, some studies have proposed the introduction of heterogeneous information networks (heterogeneous information networks) into recommendation systems. The heterogeneous information network comprises a plurality of types of nodes, and edges between the different types of nodes comprise different types of semantics. Common examples are bibliographic information networks, social media networks, protein networks, and the like. In recent years, a meta-path concept has been proposed for representing rich semantic information in a heterogeneous information network, a meta-path being a sequence of nodes of different types, each meta-path recording the type of an edge passed from an initial node to a last node and the node type. Different meta-paths contain different semantic information, so that the rich semantic information is introduced into the recommendation system, and the problems of data sparseness, cold start, over recommendation and the like of the recommendation system are hopefully solved.
The meta path is introduced into a propulsion system, and because a plurality of meta paths exist in a heterogeneous information network, how to effectively fuse the plurality of meta paths is a key for utilizing the meta path to extract consistent information from the meta paths for applying to downstream tasks. In the existing method, different weights are distributed to different paths, and then meta-path semantics are respectively fused to a regularization term, a prediction function or a user/commodity potential representation module. Although good effect is achieved, the methods only perform simple fusion on different meta-paths through the weight coefficients, and the expression capability of the model on different meta-path semantics is limited. In addition, little work is needed to study how to perform node embedding learning by using meta-path semantics and node content/attribute information simultaneously.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a recommendation method and a recommendation system fusing meta-path semantics, which can fuse semantic information contained in different meta-paths to obtain consistent semantic information and apply the semantic information to an embedded learning or recommendation strategy of a user/commodity.
In order to achieve the above object, the present invention provides a recommendation method fusing meta-path semantics, comprising: calculating a relative importance bipartite graph of the user/commodity based on a heterogeneous information network and a user/commodity scoring data set; guiding information propagation between user/commodity nodes by using the relative importance bipartite graph to obtain embedded expression of the user/commodity nodes; and transforming the embedded expression of the user/commodity node to obtain a final embedded expression of the user/commodity node for recommendation, and inputting the final embedded expression of the user/commodity node into a recommendation model to obtain the prediction score of the user/commodity.
In order to achieve the above object, the present invention further provides a recommendation system fusing meta-path semantics, including: the meta-path early fusion module is used for calculating a relative importance bipartite graph of the user/commodity based on a heterogeneous information network and a user/commodity scoring data set; the meta-path semantic embedding module is used for guiding information propagation between user/commodity nodes by utilizing the relative importance bipartite graph to obtain embedded expression of the user/commodity nodes; and the recommendation prediction module is used for transforming the embedded expression of the user/commodity node to obtain a final embedded expression of the user/commodity node for recommendation, and inputting the final embedded expression of the user/commodity node into a recommendation model to obtain the prediction score of the user/commodity.
According to the scheme, the invention has the advantages that: carrying out early fusion on the similarity based on different meta-paths to obtain consistent semantics, and directly applying the consistent semantics to the expression learning process of users/articles; by using an information propagation mechanism and an attention model for reference, the semantics based on the meta path is innovatively used as the attention score between the nodes, and the learned embedding can simultaneously reserve the content information of the nodes and the topological structure information between the nodes. In addition, the invention provides a wider recommendation model, a user can design a specific implementation mode of each module by himself, for example, a similarity matrix based on the meta-path is calculated in different modes, early fusion of a plurality of meta-paths is carried out by adopting different structures, and the model is simple and easy to build. Meanwhile, compared with the existing method, the actual recommendation effect of the method is better.
Drawings
FIG. 1 is a schematic diagram of a framework of a recommendation system fusing meta-path semantics according to the present invention.
FIG. 2 is a block diagram of the meta-path early fusion module according to the present invention.
FIG. 3 is a block diagram of a meta-path semantic embedding module according to the present invention.
Wherein, the reference numbers:
1: meta path
2: bipartite graph of relative importance
3: similarity matrix
4: previous round of embedded expressions
5: community information
6: embedded representation of the current wheel
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict.
Referring to fig. 1, the recommendation method fusing meta-path semantics provided by the present invention mainly includes three steps, that is, a bipartite graph of relative importance of users/goods is calculated based on a heterogeneous information network and a user/goods scoring dataset; guiding information propagation between user/commodity nodes by using the relative importance bipartite graph to obtain embedded expression of the user/commodity nodes; and transforming the embedded expression of the user/commodity node to obtain a final embedded expression of the user/commodity node for recommendation, and inputting the final embedded expression of the user/commodity node into a recommendation model to obtain the prediction score of the user/commodity. The recommendation system for fusing meta-path semantics provided by the invention mainly comprises three modules, namely a meta-path early fusion module, a meta-path semantics embedding module and a recommendation prediction module (recommendation), wherein the three modules respectively correspond to the three steps, and the functions of the modules are specifically described below.
In the embodiment of the present invention, let the user set be U, the commodity set be I, and the score value set be Y, where | U | ═ m, | I | ═ n. For simplicity, scoring data is considered here, giving scoring records of ═ u, v, yuv|u∈U,v∈I,yuvIs belonged to Y }. Given a heterogeneous information network G ═ V, E, where the set of node types is a, the edge type is R, and the meta-path is a path template defined over the node types and the edge types, which can be represented by the node types that pass: p ═ A1A2...Al+1). Definition of
Figure BDA0002947137190000045
Is node type AiAnd node type AjAn adjacency matrix in between. According to different application scenes, an L-shaped meta-path template P is artificially designed1,P2,...,PL. The recommendation system fusing meta-path semantics mainly comprises the following three modules:
early fusion module for one-element path
Referring to fig. 2, the meta-path early fusion module mainly calculates a bipartite graph of relative importance of users/commodities based on a heterogeneous information network and a user/commodity scoring dataset; and adaptively fusing the preprocessed similarity matrixes with different semantics.
Firstly, based on a given heterogeneous information network and L meta paths, preprocessing data and calculating a similarity matrix of user/commodity nodes under different meta path semantics. The method for calculating the similarity comprises a plurality of specific methods, the PathCount similarity is adopted in the invention, and the specific calculation method comprises the following steps: along the element path, the adjacency matrix between every two adjacent types of nodes
Figure BDA0002947137190000046
The multiplication is performed successively. The similarity is proportional to the total number of paths between the initial node and the tail node according to given semantics, and the user/commodity similarity matrix obtained under the first element path is M(l). An auxiliary bipartite graph BG (U, I, E, R) is defined, nodes of the auxiliary bipartite graph are two types of nodes of users and commodities, and only the user nodes are connected with the user nodes. The edge connected between the user node u and the commodity node v has a feature vector ruvDefined as:
Figure BDA0002947137190000041
wherein, in the formula (1)
Figure BDA0002947137190000042
Is the similarity between the u-th user and the v-th commodity under the ith meta-path semantic meaning. In the invention, r isuvConsider a multi-dimensional representation of a complex relationship between user u and item v. The edge set E of the auxiliary bipartite graph is defined as:
Figure BDA0002947137190000043
on the auxiliary bipartite graph, a neighbor set of a vertex i is defined as: n (I) ═ { j | (I, j) ∈ E }, I ∈ U £ I.
The meta-path early fusion module combines the feature vector r on the upper side of the auxiliary bipartite graphuvAnd mapping the similarity into the relative importance between the user node u and the commodity node v, so that the complex and inconsistent similarity under a plurality of meta paths is converted into the similarity under the consistent semantic meaning. Further, the method can be used for preparing a novel materialIn particular, the invention adopts a Multi-layer perceptron (MLP) to convert the feature vector r of the edgeuvMapping to a scalar euvAnd carrying out normalization processing by a Softmax function:
euv＝MLP(ruv) (2)
Figure BDA0002947137190000044
Figure BDA0002947137190000051
if the value calculated according to the formula (2) is only used as the relative importance between the user node u and the commodity node v, the importance of the user node u to the commodity node v is equal to the relative importance of the commodity node v to the user node u. In fact, the respective neighbor nodes and the number of the neighbor nodes of the user node u and the commodity node v are different, so the importance of the user node u to the commodity node v and the relative importance of the commodity node v to the user node u are also unequal, and the scalar e is further expressed by the formulas (3) and (4) in the inventionuvCarrying out normalization processing by a softmax function, and obtaining alphauvDefined as the relative importance of user node u to commodity node v, and αvuDefined as the relative importance of the commodity node v to the user node u.
All alpha is converted intouvAnd alphavuPut together and construct a matrix
Figure BDA0002947137190000052
Wherein alpha isU→IComposed of the relative importance of all users to the goods, αI→UConsisting of the relative importance of all the goods to the user. This matrix α can be viewed as a weighted adjacency matrix for the auxiliary bipartite graph, which is referred to herein as the relative importance bipartite graph. The relative importance bipartite graph alpha will be used in the next meta-path semantic embedding module to guide the embedded update of the user/good.
(II) meta-path semantic embedding module
Referring to fig. 3, the meta-path semantic embedding module mainly performs embedding learning of user/commodity nodes based on the relative importance bipartite graph α output by the meta-path early fusion module, and the learned user/commodity nodes can simultaneously maintain the content features of the nodes and the topological structure between the nodes (i.e., the topological relationship of the graph corresponding to the relative importance bipartite graph α).
Setting the original characteristics of the user nodes as
Figure BDA0002947137190000053
Wherein each row represents a p-dimensional original feature of a user, and the original features of the commodity nodes are
Figure BDA0002947137190000054
Where each row represents a q-dimensional primitive feature of a good. Firstly, the original characteristics of the user nodes are transformed through a characteristic
Figure BDA0002947137190000055
Raw characteristics of commodity nodes
Figure BDA0002947137190000056
Converting the data into the same space, and splicing the data to obtain an initial characteristic matrix of user/commodity nodes
Figure BDA0002947137190000057
Figure BDA0002947137190000058
Wherein, in the formula (5)
Figure BDA0002947137190000059
And the transformation matrixes respectively represent the characteristics of the user/commodity nodes to be learned. The meta-path semantic embedding module carries out the embedded updating of the user/commodity nodes through an information propagation mechanism,in order to fuse the meta-path semantics and the node characteristics, the invention designs a bipartite graph alpha which guides information propagation by using relative importance output by a meta-path early fusion module. That is, the relative importance bipartite graph α acts as a mechanism of attention, directing the propagation of information between user/commodity nodes. The information propagation is supposed to carry out Γ round iteration, each round of iteration can be divided into two processes, and it should be noted here that the obtained initial feature matrix can be used as the embedded expression of the user/commodity node of the first round. For example, for the t-th round in the iteration of the Γ -th round, first, the present invention uses an Aggregator (Aggregator) to apply the current characteristics of the neighbor nodes of each node i (user/commodity)
Figure BDA0002947137190000061
j e N (i) are aggregated to obtain the community information of each node i after aggregation
Figure BDA0002947137190000062
Wherein:
Figure BDA0002947137190000063
obtaining community information of each node i
Figure BDA0002947137190000064
Then, the previous round of embedded expression of each node i is performed
Figure BDA0002947137190000065
And corresponding community information
Figure BDA0002947137190000066
A Filter (Filter) is used to Filter information once and obtain the updated embedded expression of each node i of the current round
Figure BDA0002947137190000067
Figure BDA0002947137190000068
The formula (6) has the function that the semantics based on the meta path and the content information of the nodes are fused through an aggregator, and the information propagation of the nodes is guided by skillfully utilizing the meta path semantic information by means of an attention mechanism, so that the learned nodes simultaneously keep the topological relation between the content information of the nodes and the nodes.
(III) recommendation prediction module
The recommendation prediction module mainly has the function of embedding expression H of updated nodes (users/commodities) obtained by the meta-path semantic embedding moduleΓAnd (after gamma round iteration) performing a downstream prediction task of the model. Based on the embedding of user/commodity nodes, the framework provided by the invention can conveniently carry out classification, regression and other prediction tasks, and the invention focuses on scoring data (prediction scores)
Figure BDA00029471371900000613
) The method is used for evaluating the relative importance between the user and the commodity, and two recommendation prediction models are designed correspondingly.
First, the embedded expression H of the node (user/commodity)ΓAnd transforming through a full connection layer to obtain the final embedded expression of the recommended user/commodity:
Figure BDA0002947137190000069
i belongs to U and I. Based on this, the invention provides two recommendation prediction models: linear models and bilinear models. The linear model weights the vector obtained by splicing the final embedded expression of the user and the final embedded expression of the commodity to obtain the prediction score
Figure BDA00029471371900000610
The weighting coefficient is
Figure BDA00029471371900000611
The bilinear model uses a symmetric positive definite matrix to determine the maximum of the userThe final embedded expression and the final embedded expression of the commodity are mapped through a bilinear function to obtain a prediction score
Figure BDA00029471371900000612
Compared with a linear model, the bilinear model introduces secondary interaction and has more flexibility. Wherein,
linear model:
Figure BDA0002947137190000071
bilinear model:
Figure BDA0002947137190000072
in addition, the score data (prediction score) is targeted
Figure BDA0002947137190000073
) The invention constructs the regression loss
Figure BDA0002947137190000074
And optimizing the model by a random gradient descent method.
The invention has performed experiments on two real data sets MovieLens-1M and Yelp. The evaluation indices used were the common RMSE (root mean square error) and MAE (mean absolute error). For any user, the calculation formulas of the RMSE and the MAE are respectively:
Figure BDA0002947137190000075
Figure BDA0002947137190000076
wherein m is the number of test samples, yiAnd
Figure BDA0002947137190000077
respectively scoring and predicting the user's true evaluation of the ith test sampleThe evaluation score of the user on the ith test sample.
The following three major types of comparison methods were compared: (1) using other data except user/commodity scoring data in the abnormal picture as the characteristics of user or commodity, such as FM [11] and GCMC [12 ]; (2) method based on meta-path: such as FMG [13] and HAN [14 ]; (3) methods based on knowledge-map embedding, such as CKE [15] and KGAT [16 ].
The MovieLens-1M dataset used (https:// group. org/datasets/MovieLens /) contains different node types as follows: user (user, abbreviated as U), movie (movie, abbreviated as M), gender (Gd), occupation (Ocp), and movie type (type, abbreviated as T). The relationship contained is as follows: UM (user watching movie), UOcp (user belonging to a certain occupation), UGd (user belonging to a certain gender), MT (movie belonging to a certain genre). The artificially designed meta-path is: UM, UMUM, UMTM, UMTMUM, UOcpUM, UGdUM.
The Yelp dataset (https:// www.yelp.com/dataset challenge) used contained different node types as follows: user (user, abbreviation U), merchant (business, abbreviation B), quartic prize (comment, abbreviation Comp), city (city, abbreviation Ci), type (category, abbreviation Cat). The following relationships are included: the user evaluates the merchant (UB), the user and the user are friends (UU), the user conducts a quartic prize (UComp), the merchant belongs to a certain city (BCi), and the merchant belongs to a certain type (BCat). The following meta-paths were designed manually: UB, UBUBUB, UUB, UCompUB, UBCiB, UBCatB, UBCatBB.
And taking the unique hot codes of the user ID and the commodity ID as original input features, taking 128 as a feature dimension k, and taking 24 as a final feature dimension d. The total iteration round number Γ takes 1. An Adam optimizer was used, and the learning rate was set to 0.001. Experiments were performed under three different training/testing settings, the number of training samples/the number of testing samples being: 40%/60%, 60%/40%, 80%/20%. The results of the experiments on the two data sets are shown in table 1 and table 2, respectively.
TABLE 1
Figure BDA0002947137190000081
Table 1 shows the results of the experiments on the MovieLens-1M data set. Where ↓ represents the smaller the value, the better the model performance. Bolded indicates the best experimental results, underlined indicates the second best experimental results in the prior art.
TABLE 2
Figure BDA0002947137190000082
Table 2 shows the results of the experiments on the Yelp data set. Where ↓ represents the smaller the value, the better the model performance. The bold indicates the best experimental results of the present invention, and the underline indicates the suboptimal experimental results in the prior art method. Since the Yelp data set is very sparse at training set ratios of 40% and 60%, the two types of comparison are not applicable to the knowledge-based map embedding method, and are not compared, and the tables are marked as '-'.
As can be seen from the experimental results in tables 1 and 2, the present invention achieves superior performance on both real data sets compared to the prior art method.
In conclusion, the method carries out early fusion on the similarity based on different meta-paths to obtain consistent semantics, and is directly applied to the expression learning process of users/articles; by using an information propagation mechanism and an attention model for reference, the semantics based on the meta path is innovatively used as the attention score between the nodes, and the learned embedding can simultaneously reserve the content information of the nodes and the topological structure information between the nodes. In addition, the invention provides a wider recommendation model, a user can design a specific implementation mode of each module by himself, for example, a similarity matrix based on the meta-path is calculated in different modes, early fusion of a plurality of meta-paths is carried out by adopting different structures, and the model is simple and easy to build. Meanwhile, compared with the existing method, the actual recommendation effect of the method is better.

Claims (10)

1. A recommendation method fusing meta-path semantics is characterized by comprising the following steps:
calculating a relative importance bipartite graph of the user/commodity nodes based on a heterogeneous information network and a grading data set of the user/commodity;
guiding information propagation between user/commodity nodes by using the relative importance bipartite graph to obtain embedded expression of the user/commodity nodes;
and transforming the embedded expression of the user/commodity node to obtain a final embedded expression of the user/commodity node for recommendation, and inputting the final embedded expression of the user/commodity node into a recommendation model to obtain the prediction score of the user/commodity.
2. The recommendation method fusing meta-path semantics as claimed in claim 1, wherein the calculating the bipartite graph of relative importance of the user/commodity nodes based on a heterogeneous information network and a user/commodity scoring dataset comprises:
calculating similarity matrix M of the user/commodity node in different element paths(l)；
Calculating a feature vector r of edges between the user/commodity nodesuv；
Adopting a multilayer perceptron to convert the feature vector r of the edgeuvMapping to a scalar euvCarrying out normalization processing by a Softmax function to construct a bipartite graph alpha of the relative importance of the user/commodity;
wherein,
Figure FDA0002947137180000011
euv＝MLP(ruv)，
Figure FDA0002947137180000012
Figure FDA0002947137180000013
the similarity between the user node u and the commodity node v along the ith element path is shown.
3. The recommender fusing meta-path semantics as claimed in claim 2Method, characterized in that said scalar euvThe normalization processing by the Softmax function comprises the following steps:
Figure FDA0002947137180000014
Figure FDA0002947137180000015
wherein alpha isuvDefined as the relative importance of the user node u to the commodity node v, αvuDefined as the relative importance of the commodity node v to the user node u.
4. The recommendation method fusing meta-path semantics as claimed in claim 1, wherein the guiding information propagation between user/commodity nodes by using the relative importance bipartite graph to obtain the embedded expression of the user/commodity nodes comprises:
the original characteristics of the user nodes are combined
Figure FDA0002947137180000016
The original characteristics of the commodity node are
Figure FDA0002947137180000017
After converting to the same space, splicing to obtain the initial characteristic matrix of the user/commodity node
Figure FDA0002947137180000021
The current characteristics of the neighbor nodes of the user/commodity nodes
Figure FDA0002947137180000022
Aggregating through an aggregator to obtain the community information of the user/commodity node
Figure FDA0002947137180000023
Expressing the embedding of the user/commodity node of the previous round
Figure FDA0002947137180000024
And the community information
Figure FDA0002947137180000025
Obtaining the embedded expression of the user/commodity node of the current round through a filter
Figure FDA0002947137180000026
Wherein, a plurality of rounds of information transmission are carried out,
Figure FDA0002947137180000027
as a first round of embedded expressions for the user/commodity node,
Figure FDA0002947137180000028
transformation matrices respectively representing the user/commodity node characteristics to be learned,
Figure FDA0002947137180000029
Figure FDA00029471371800000210
5. the method of recommendation fusing meta-path semantics of claim 1, wherein the recommendation model comprises: a linear model or a bilinear model.
6. A recommendation system fusing meta-path semantics, comprising:
the meta-path early fusion module is used for calculating a relative importance bipartite graph of the user/commodity node based on a heterogeneous information network and a grading data set of the user/commodity;
the meta-path semantic embedding module is used for guiding information propagation between user/commodity nodes by utilizing the relative importance bipartite graph to obtain embedded expression of the user/commodity nodes;
and the recommendation prediction module is used for transforming the embedded expression of the user/commodity node to obtain a final embedded expression of the user/commodity node for recommendation, and inputting the final embedded expression of the user/commodity node into a recommendation model to obtain the prediction score of the user/commodity.
7. The meta-path semantic fused recommendation system of claim 6 wherein said meta-path early fusion module comprises:
a similarity matrix calculation submodule for calculating a similarity matrix M of the user/commodity node in different element paths(l)；
A feature vector calculation submodule for calculating a feature vector r of an edge between the user/commodity nodesuv；
Constructing a sub-module of the relative importance bipartite graph, which is used for adopting a multilayer perceptron to convert the feature vector r of the edgeuvMapping to a scalar euvCarrying out normalization processing by a Softmax function to construct a bipartite graph alpha of the relative importance of the user/commodity;
wherein,
Figure FDA00029471371800000211
euv＝MLP(ruv)，
Figure FDA00029471371800000212
Figure FDA00029471371800000213
and the similarity between the user node u of the ith element path and the commodity node v.
8. The meta-path semantic fused recommendation system of claim 7, further comprisingCharacterized in that said scalar euvThe normalization processing by the Softmax function comprises the following steps:
Figure FDA0002947137180000031
Figure FDA0002947137180000032
wherein alpha isuvDefined as the relative importance of the user node u to the commodity node v, αvuDefined as the relative importance of the commodity node v to the user node u.
9. The system for recommending fusing meta-path semantics of claim 6, wherein said meta-path semantics embedding module comprises:
an initial feature matrix splicing submodule for splicing the original features of the user nodes
Figure FDA0002947137180000033
The original characteristics of the commodity node are
Figure FDA0002947137180000034
After converting to the same space, splicing to obtain the initial characteristic matrix of the user/commodity node
Figure FDA0002947137180000035
An aggregation submodule for aggregating the current characteristics of the neighbor nodes of the user/commodity nodes
Figure FDA0002947137180000036
Aggregating through an aggregator to obtain the community information of the user/commodity node
Figure FDA0002947137180000037
A filtering submodule for expressing the embedded user/commodity node in the previous round
Figure FDA0002947137180000038
And the community information
Figure FDA0002947137180000039
Obtaining the embedded expression h of the user/commodity node of the current round through a filterit；
Wherein, a plurality of rounds of information transmission are carried out,
Figure FDA00029471371800000310
as a first round of embedded expressions for the user/commodity node,
Figure FDA00029471371800000311
transformation matrices respectively representing the user/commodity node characteristics to be learned,
Figure FDA00029471371800000312
Figure FDA00029471371800000313
10. the system for recommendation fusing meta-path semantics of claim 6, wherein the recommendation model comprises: a linear model or a bilinear model.
CN202110198542.4A 2021-02-22 2021-02-22 Recommendation method and system for fusion element path semantics Active CN112861006B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110198542.4A CN112861006B (en) 2021-02-22 2021-02-22 Recommendation method and system for fusion element path semantics

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110198542.4A CN112861006B (en) 2021-02-22 2021-02-22 Recommendation method and system for fusion element path semantics

Publications (2)

Publication Number Publication Date
CN112861006A true CN112861006A (en) 2021-05-28
CN112861006B CN112861006B (en) 2023-06-23

Family

ID=75988526

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110198542.4A Active CN112861006B (en) 2021-02-22 2021-02-22 Recommendation method and system for fusion element path semantics

Country Status (1)

Country Link
CN (1) CN112861006B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113836425A (en) * 2021-09-29 2021-12-24 平安科技（深圳）有限公司 Movie recommendation method, device and equipment based on heteromorphic image attention network
CN114840777A (en) * 2022-07-04 2022-08-02 杭州城市大脑有限公司 Multi-dimensional endowment service recommendation method and device and electronic equipment
CN114896514A (en) * 2022-07-14 2022-08-12 西安电子科技大学 Web API label recommendation method based on graph neural network
CN115795177A (en) * 2022-11-25 2023-03-14 人民网股份有限公司 Social media content recommendation method and device

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120030159A1 (en) * 2010-07-30 2012-02-02 Gravity Research & Development Kft. Recommender Systems and Methods
CN111310045A (en) * 2020-02-16 2020-06-19 重庆邮电大学 Network-embedded movie recommendation method based on meta-path
CN112148931A (en) * 2020-09-29 2020-12-29 河北工业大学 Meta path learning method for high-order abnormal picture classification
CN112182424A (en) * 2020-11-11 2021-01-05 重庆邮电大学 Social recommendation method based on integration of heterogeneous information and isomorphic information networks
CN112364245A (en) * 2020-11-20 2021-02-12 浙江工业大学 Top-K movie recommendation method based on heterogeneous information network embedding

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120030159A1 (en) * 2010-07-30 2012-02-02 Gravity Research & Development Kft. Recommender Systems and Methods
CN111310045A (en) * 2020-02-16 2020-06-19 重庆邮电大学 Network-embedded movie recommendation method based on meta-path
CN112148931A (en) * 2020-09-29 2020-12-29 河北工业大学 Meta path learning method for high-order abnormal picture classification
CN112182424A (en) * 2020-11-11 2021-01-05 重庆邮电大学 Social recommendation method based on integration of heterogeneous information and isomorphic information networks
CN112364245A (en) * 2020-11-20 2021-02-12 浙江工业大学 Top-K movie recommendation method based on heterogeneous information network embedding

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113836425A (en) * 2021-09-29 2021-12-24 平安科技（深圳）有限公司 Movie recommendation method, device and equipment based on heteromorphic image attention network
CN114840777A (en) * 2022-07-04 2022-08-02 杭州城市大脑有限公司 Multi-dimensional endowment service recommendation method and device and electronic equipment
CN114840777B (en) * 2022-07-04 2022-09-27 杭州城市大脑有限公司 Multi-dimensional endowment service recommendation method and device and electronic equipment
CN114896514A (en) * 2022-07-14 2022-08-12 西安电子科技大学 Web API label recommendation method based on graph neural network
CN114896514B (en) * 2022-07-14 2022-09-30 西安电子科技大学 Web API label recommendation method based on graph neural network
CN115795177A (en) * 2022-11-25 2023-03-14 人民网股份有限公司 Social media content recommendation method and device
CN115795177B (en) * 2022-11-25 2023-04-28 人民网股份有限公司 Social media content recommendation method and device

Also Published As

Publication number Publication date
CN112861006B (en) 2023-06-23

Similar Documents

Publication Publication Date Title
CN112861006A (en) Recommendation method and system fusing meta-path semantics
CN113362131A (en) Intelligent commodity recommendation method based on map model and integrating knowledge map and user interaction
CN113918832A (en) Graph convolution collaborative filtering recommendation system based on social relationship
CN115860880B (en) Personalized commodity recommendation method and system based on multi-layer heterogeneous graph convolution model
Shao et al. FairCF: Fairness-aware collaborative filtering
CN113918833A (en) Product recommendation method realized through graph convolution collaborative filtering of social network relationship
CN113918834A (en) Graph convolution collaborative filtering recommendation method fusing social relations
CN115408605A (en) Neural network recommendation method and system based on side information and attention mechanism
WO2024061073A1 (en) Multimedia information generation method and apparatus, and computer-readable storage medium
CN109800424A (en) It is a kind of based on improving matrix decomposition and the recommended method across channel convolutional neural networks
CN113342994A (en) Recommendation system based on non-sampling cooperative knowledge graph network
CN116304289A (en) Information chain recommendation method and device for supply chain based on graphic neural network
Liang et al. The graph embedded topic model
Hoang et al. Improving Graph Convolutional Networks with Transformer Layer in social-based items recommendation
CN115564013B (en) Method for improving learning representation capability of network representation, model training method and system
CN118246939A (en) Method and system for processing DeepFM model browsing data based on attention mechanism
Fu et al. Preference-aware heterogeneous graph neural networks for recommendation
Lu et al. BERT-RS: A neural personalized recommender system with BERT
Sun et al. DeepPRFM: Pairwise Ranking Factorization Machine Based on Deep Neural Network Enhancement
Chen et al. A context-aware collaborative filtering recommender system based on GCNs
CN115545467B (en) Risk commodity identification model based on graphic neural network
Pang et al. Leveraging Knowledge Graph and Self-Attention with Residual Block for Paper Recommendation
Zhou et al. Unexp-DIN: Unexpected Deep Interest Network for Recommendation
Sun User behavior analysis of cloud service based on graph neural network research on personalized recommendation algorithm
Zhang et al. Towards Better Top-K Recommendation with Attention-Aware Collaborative Matrix Factorization

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