CN114820139A - Multi-user recommendation system based on knowledge graph path reasoning - Google Patents

Multi-user recommendation system based on knowledge graph path reasoning Download PDF

Info

Publication number
CN114820139A
CN114820139A CN202210583484.1A CN202210583484A CN114820139A CN 114820139 A CN114820139 A CN 114820139A CN 202210583484 A CN202210583484 A CN 202210583484A CN 114820139 A CN114820139 A CN 114820139A
Authority
CN
China
Prior art keywords
path
user
graph
module
knowledge graph
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
CN202210583484.1A
Other languages
Chinese (zh)
Other versions
CN114820139B (en
Inventor
王丽平
杨正益
柳玲
危枫
周魏
文俊浩
郭向星
程旺鑫
杨佳佳
朱磊
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.)
Chongqing University
Original Assignee
Chongqing University
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 Chongqing University filed Critical Chongqing University
Priority to CN202210583484.1A priority Critical patent/CN114820139B/en
Publication of CN114820139A publication Critical patent/CN114820139A/en
Application granted granted Critical
Publication of CN114820139B publication Critical patent/CN114820139B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • 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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • 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
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • General Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Evolutionary Computation (AREA)
  • Development Economics (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Databases & Information Systems (AREA)
  • Animal Behavior & Ethology (AREA)
  • Game Theory and Decision Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a multi-user recommendation system based on knowledge graph path inference, which comprises a knowledge graph construction module, a path inference module and a grading prediction module; the knowledge graph construction module acquires user interaction historical data, constructs a knowledge graph G and transmits the knowledge graph G to the path reasoning module; the path reasoning module generates a plurality of relation paths between the users and the target project according to the knowledge graph G and transmits the relation paths to the grading prediction module; and the scoring prediction module carries out evaluation prediction on the received relation paths and outputs a project recommendation list to the user according to the evaluation prediction result. According to the method, more abundant user interest latent semantic information is mined by constructing the knowledge graph of the users and items, the potential common preference of a plurality of users is predicted by combining and reasoning a plurality of paths on the knowledge graph, and the recommendation list with maximized interest preference and diversification is provided while different user preference degrees are distinguished by using pooling aggregation operation with attention mechanism among different paths.

Description

Multi-user recommendation system based on knowledge graph path reasoning
Technical Field
The invention relates to the field of user recommendation, in particular to a multi-user recommendation system based on knowledge graph path reasoning.
Background
Most of recommendation technologies studied in academia and industry are personalized recommendation systems for a single user, however, recommended objects are often groups of more than one person, and then group recommendation technologies are born. However, there are naturally occurring groups such as families, classes, companies, etc., and the interest relevance between the groups may be small, and in some cases, the groups are limited to purchase resources only through the same device ID or the same login account. For these multiple users, the conventional group recommendation system is not applicable, and a multi-user recommendation system is derived for this purpose.
Typical application scenarios of the multi-user recommendation system are as follows: 1) the IPTV field: a plurality of family members of the household television can only watch the television on line through the same set top box IP; 2) the video mobile client: a plurality of clients share the same member account to log in the fancy art and the Tencent video to select respective interested film and television resources; 3) e, E-commerce platform: in order to enjoy discount service, family members share the same member account number to log in an e-commerce platform to select and purchase commodities and the like on line. For the users, the interests of the internal members of the users are greatly different, however, due to natural factors, a multi-user group is constructed, and the commodities are purchased through the same account or the same equipment IP. And recommending contents meeting the interests of members in the groups for the groups, which is called a multi-user recommendation system.
How to construct a multi-user recommendation system enables the user to find a recommendation system which not only meets the requirements of a plurality of users but also embodies personalized recommendation in information-overloaded mass resources through the same equipment IP or the same account number, and the following problems need to be solved:
1) maximizing the community interest preferences among the members;
2) the recommendation results are required to be diversified to meet individual requirements among members;
3) the social relationship between groups is lacked, and how to utilize the interaction history of the existing users and items to mine more effective information;
disclosure of Invention
The invention aims to provide a knowledge graph path inference based multi-user recommendation system, which comprises a knowledge graph construction module, a path inference module and a grading prediction module;
the knowledge graph construction module acquires user interaction historical data, constructs a knowledge graph G and transmits the knowledge graph G to the path reasoning module;
the step of constructing the knowledge-graph G comprises the following steps:
a) building item associated directed graph G 1 (ii) a The item association directed graph G 1 The entity of (1) comprises an interactive item and a candidate item, and the edge set comprises an incidence relation between the entity and the entity;
item associated directed graph G 1 ＝{(h,r,t)|h,t∈I 1 ,r∈R 1 Obtaining through modeling of associated data between the candidate items and the interacted items;
wherein the tuple (h, r, t) indicates that a relationship r exists between the head node h and the tail node t; I.C. A 1 All candidate item sets are selected; r 1 Constructing an interaction directed graph G of users and items for the relation set b) 2 (ii) a The interaction directed graph G 2 The entities of (1) include users and items; the interaction directed graph G 2 If the user and the project have interaction, an edge exists between the user and the project;
user interaction directed graph G with items 2 ＝{(u,act,i)|u∈U,i∈I 2 ,act∈R 2 The method is obtained by modeling of interaction records of users and projects;
wherein, the tuple (u, act, i) represents that the interaction relation act exists between the user u and the item i;
Figure BDA0003662626020000021
and
Figure BDA0003662626020000022
respectively representing users, item sets, M and N respectively representing the number of users and the number of items,
Figure BDA0003662626020000023
R 2 a set of interactions is represented.
c) Associating items with directed graph G 1 And user interaction directed graph G with items 2 And fusing to obtain a knowledge graph G.
The knowledge graph G { (h, R, t) | h, t ∈ E, R ∈ R }, wherein the set E ═ U ≡ I { (h, R, t) | h, t ∈ E, R ∈ R }, and the set E { [ U ] } I } 1 The set R ═ R 1 ∪R 2 . The path reasoning module generates a plurality of relation paths between the users and the target project according to the knowledge graph G and transmits the relation paths to the grading prediction module;
the step of the path inference module generating a relationship path between the plurality of users and the target item comprises:
1) recording paths between users and the items on the knowledge graph G as sequences of entities and relations
Figure BDA0003662626020000024
Wherein e is 1 ＝u,e L ＝p；(e l ,r l ,e l+1 ) Is the ith tuple on path p, L-1 represents the number of all tuples on path p;
2) for a given relationship path p on the knowledge-graph G k Said path inference module links the relationship path p k Each entity in (1) is partitioned into a corresponding type and value, and the type and value are initialized into two vectors which are respectively recorded as
Figure BDA0003662626020000025
d represents the dimension size; k is initially 1;
3) on the path p k In which vectors r of the same dimension are embedded l Updating relationship path p' k Is [ e ] 1 ,r 1 ,e 2 ,...,e l-1 ,r l-1 ]Relation path p' k Each element in the list represents an entity or a type of relationship;
4) mining a relationship path p 'using a recurrent neural network model stored on a path inference module' k Thereby generating an encoded path p k ；
The circulating neural network model is an LSTM neural network model and comprises an input layer, a hiding layer and an output layer; the hidden layer comprises an input gate, a forgetting gate and an output gate;
the input of the recurrent neural network model comprises entity type, entity value and relation vector, and the output is the encoded path.
Input vector x of the recurrent neural network model l-1 Output vector h L Respectively satisfy the following formula:
Figure BDA0003662626020000031
z l ＝tanh(W 1 x l +W h h l-1 +b z ) (2)
f l ＝σ(W f x l +W h h l-1 +b f ) (3)
i l ＝σ(W i x l +W h h l-1 +b i ) (4)
o l ＝σ(W o x l +W h h l-1 +b o ) (5)
c l ＝f l ⊙c l-1 +i l ⊙z l (6)
h l ＝o l ⊙tanh(c l ) (7)
in the formula, c l ∈R d ,z l ∈R d Respectively representing the unit storage state vector and the information conversion module, and d' represents the number of the hidden units; i.e. i l 、o l And f l Respectively representing input, output and forgetting gate, W z 、W i 、W f 、W o ∈R d'×3d ,W h ∈R d'×d' Representing a mapping coefficient matrix; b z 、b i 、b f 、b o Is a bias vector; σ (.) is an activation function, an indicates the element-level product between two vectors; e.g. of the type l-1 、e′ l-1 、r l-1 Respectively representing entity types, entity values and relationship vectors.
5) Let k be k +1 and return to step 2) until the encoded path set P (u, i) is obtained as { P } 1 ,p 2 ,...,p K }。
And the scoring prediction module carries out evaluation prediction on the received relation paths and outputs a project recommendation list to the user according to the evaluation prediction result.
The items in the item recommendation list are predicted according to scores
Figure BDA0003662626020000032
The Top N items in the descending order (Top-N recommendation).
The step of the scoring prediction module for evaluating and predicting the received relationship paths comprises the following steps:
I) establishing a prediction function, namely:
Figure BDA0003662626020000041
in the formula (I), the compound is shown in the specification,
Figure BDA0003662626020000042
to evaluate the prediction results, f θ Representing a model with a parameter θ, which indicates whether the user u interacts with the item i is inferred from the connectivity ρ (u, i);
II) calculating a state prediction value, namely:
Figure BDA0003662626020000043
in the formula, W 1 ，W 2 Respectively representing coefficient weights of a 1 st layer full connection layer and a 2 nd layer full connection layer; relu denotes an activation function; τ ═ (u, act, i) denotes a tuple; s (τ | p) k )＝s k And representing the predicted score of the k-th path from the node u to the node i.
III) projecting the state predicted value to a prediction function to obtain an evaluation prediction result
Figure BDA0003662626020000044
Namely:
Figure BDA0003662626020000045
where σ () is the activation function;
wherein the parameter g(s) 1 ,s 2 ,...,s K ) As follows:
Figure BDA0003662626020000046
in which gamma is controlMaking a hyperparameter for each exponential weight; { s 1 ,s 2 ,...,s K And expressing the respective prediction scores of the total K paths from the node u to the node i. s k And representing the predicted score of the k-th path from the node u to the node i.
The technical effect of the invention is undoubted, and the invention provides a knowledge graph path inference-based multi-user recommendation system, which is characterized in that more abundant user interest latent semantic information is mined by constructing a knowledge graph of users and items, the potential common preference of a plurality of users is predicted by utilizing the combination inference of a plurality of paths on the knowledge graph, and a recommendation list with maximized interest preference and diversification is given by utilizing the pooling aggregation operation with attention mechanism among different paths while distinguishing the preference degrees of different users.
The method establishes a Knowledge Graph (KG) according to the historical interaction data of the user and the association information between projects, extracts a reasonable path between the user and the projects from the KG, and models the sequential dependency relationship between entities such as user projects and the like and the complex relationship between paths connecting the user and the projects. Long Short Term Memory (LSTM) networks are employed to model sequential dependencies of entities and relationships;
the invention designs the aggregation operation of the weighting pool, differentiates different contribution values of a plurality of paths between user-item interactive tuples, and has a certain attention mechanism for differentiating interest preferences among different users; compared with the traditional operation of taking the mean value of all paths, the recommendation algorithm realizes the interpretability of the path level;
the knowledge graph is applied to the multi-user recommendation method for the first time, the traditional multi-user recommendation model is expanded, a knowledge-aware path cycle neural network is provided, more candidate items which accord with the interest of the user are deduced through the historical record of the user, long-tail items are beneficially mined, and the diversity of recommendation results is improved;
besides learning the relation expression of the user and the project, the invention also extracts the complex and diverse association relation among project products. Meanwhile, the diversity of the paths is ensured, the setting of the path pooling hyperparameters is in direct proportion to the score of each path in the back propagation process, and more flexibility is provided for final prediction.
Drawings
Fig. 1 is a general flow chart of the present scheme.
Fig. 2 is a schematic diagram of the model structural components of the present solution.
FIG. 3 is a schematic diagram of a core knowledge graph path inference model.
Detailed Description
The present invention is further illustrated by the following examples, but it should not be construed that the scope of the above-described subject matter is limited to the following examples. Various substitutions and alterations can be made without departing from the technical idea of the invention and the scope of the invention is covered by the present invention according to the common technical knowledge and the conventional means in the field.
Example 1:
referring to fig. 1 to 3, a knowledge graph path inference based multi-user recommendation system includes a knowledge graph construction module, a path inference module, and a score prediction module;
the knowledge graph construction module acquires user interaction historical data, constructs a knowledge graph G and transmits the knowledge graph G to the path reasoning module;
the step of constructing the knowledge-graph G comprises the following steps:
a) building item associated directed graph G 1 (ii) a The item association directed graph G 1 The entity of (1) comprises an interactive item and a candidate item, and the edge set comprises an incidence relation between the entity and the entity;
item associated directed graph G 1 ＝{(h,r,t)|h,t∈I 1 ,r∈R 1 Obtaining through modeling of associated data between the candidate items and the interacted items;
wherein the tuple (h, r, t) indicates that a relationship r exists between the head node h and the tail node t; i is 1 All candidate item sets are selected; r 1 Is a set of relationships.
b) Building a directed graph G of user-item interactions 2 (ii) a The interaction directed graph G 2 The entities of (1) include users and items; the interaction directed graph G 2 If there is an interaction between the user and the item, there is an edge between the user and the item；
User interaction directed graph G with items 2 ＝{(u,act,i)|u∈U,i∈I 2 ,act∈R 2 The user is modeled with an interaction record of the item;
wherein, the tuple (u, act, i) represents that the interaction relation act exists between the user u and the item i;
Figure BDA0003662626020000061
and
Figure BDA0003662626020000062
respectively representing users, item sets, M and N respectively representing the number of users and the number of items,
Figure BDA0003662626020000063
R 2 a set of interactions is represented.
c) Associating items with directed graph G 1 And user interaction directed graph G with items 2 And fusing to obtain a knowledge graph G.
The knowledge graph G { (h, R, t) | h, t ∈ E, R ∈ R }, wherein the set E ═ U ≡ I { (h, R, t) | h, t ∈ E, R ∈ R }, and the set E { [ U ] } I } 1 In which the set R is R 1 ∪R 2 。。
The path reasoning module generates a plurality of relation paths between the users and the target project according to the knowledge graph G and transmits the relation paths to the grading prediction module;
the step of the path inference module generating a relationship path between the plurality of users and the target item comprises:
1) recording the path between the user and the project on the knowledge graph G as the sequence of entities and relations
Figure BDA0003662626020000064
Wherein e is 1 ＝u,e L ＝p；(e l ,r l ,e l+1 ) Is the ith tuple on path p, L-1 represents the number of all tuples on path p;
2) for a given relationship path p on the knowledge-graph G k Said path inference module links the relationship path p k Each entity in (2) is partitioned into corresponding classesType and value, and initializing the type and value into two vectors, respectively
Figure BDA0003662626020000065
d represents the dimension size; k is initially 1;
3) on the path p k In which vectors r of the same dimension are embedded l Updating relationship path p' k Is [ e ] 1 ,r 1 ,e 2 ,...,e l-1 ,r l-1 ]Relation path p' k Each element in the list represents an entity or a type of relationship;
4) mining relationship path p 'using a recurrent neural network model stored on a path inference module' k Thereby generating an encoded path p k ；
The circulating neural network model is an LSTM neural network model and comprises an input layer, a hiding layer and an output layer; the hidden layer comprises an input gate, a forgetting gate and an output gate;
the input of the recurrent neural network model comprises entity type, entity value and relation vector, and the output is the encoded path.
Input vector x of the recurrent neural network model l-1 Output vector h L Respectively satisfy the following formula:
Figure BDA0003662626020000071
z l ＝tanh(W 1 x l +W h h l-1 +b z ) (2)
f l ＝σ(W f x l +W h h l-1 +b f ) (3)
i l ＝σ(W i x l +W h h l-1 +b i ) (4)
o l ＝σ(W o x l +W h h l-1 +b o ) (5)
c l ＝f l ⊙c l-1 +i l ⊙z l (6)
h l ＝o l ⊙tanh(c l ) (7)
in the formula, c l ∈R d ,z l ∈R d Respectively representing the unit storage state vector and the information conversion module, and d' represents the number of the hidden units; i.e. i l 、o l And f l Respectively representing input, output and forgetting gate, W z 、W i 、W f 、W o ∈R d'×3d ,W h ∈R d'×d' Representing a mapping coefficient matrix; b z 、b i 、b f 、b o Is a bias vector; σ (.) is an activation function, an indicates the element-level product between two vectors; e.g. of the type l-1 、e′ l-1 、r l-1 Respectively representing entity types, entity values and relationship vectors.
5) Let k be k +1 and return to step 2) until the encoded path set P (u, i) is obtained as { P } 1 ,p 2 ,...,p K }。
And the scoring prediction module carries out evaluation prediction on the received relation paths and outputs a project recommendation list to the user according to the evaluation prediction result.
The items in the item recommendation list are predicted results according to scores
Figure BDA0003662626020000072
The Top N items in the descending order (Top-N recommendation).
The step of the scoring prediction module for evaluating and predicting the received relationship paths comprises the following steps:
I) establishing a prediction function, namely:
Figure BDA0003662626020000073
in the formula (I), the compound is shown in the specification,
Figure BDA0003662626020000074
to evaluate the predicted outcome; f. of θ A model with a parameter theta is represented, which indicates whether the user u interacts with the item i is inferred from the connectivity p (u, i).
II) calculating a state prediction value, namely:
Figure BDA0003662626020000075
in the formula, W 1 ，W 2 Respectively representing coefficient weights of a 1 st layer full connection layer and a 2 nd layer full connection layer; relu denotes an activation function; where τ ═ (u, act, i) represents a tuple; s (τ | p) k )＝s k Representing the predicted score of the k path from the node u to the node i;
III) projecting the state predicted value to a prediction function to obtain an evaluation prediction result
Figure BDA0003662626020000081
Namely:
Figure BDA0003662626020000082
where σ () is the activation function;
wherein the parameter g(s) 1 ,s 2 ,...,s K ) As follows:
Figure BDA0003662626020000083
where γ is a hyperparameter controlling each exponential weight; { s 1 ,s 2 ,...,s K And expressing the respective prediction scores of the total K paths from the node u to the node i. s k And representing the predicted score of the k-th path from the node u to the node i.
Example 2:
a multi-user recommendation system based on knowledge graph path inference comprises a knowledge graph construction module, a path inference module and a grading prediction module.
A knowledge graph construction module:
the knowledge graph construction module firstly constructs a project associated directed graph G 1 The entities are interactive items and candidate items, and the edge set comprises the incidence relation (such as same type relation and collocation relation) between the entities; secondly, constructing an interactive directed graph G of the user and the project 2 The entities are users and items, and if the users and the items have interaction, edges exist. Fusion G 1 And G 2 And obtaining a unified knowledge graph G.
Specifically, the association data between the candidate item and the interacted item is first modeled as a directed graph G 1 ＝{(h,r,t)|h,t∈I 1 ,r∈R 1 Each tuple (h, r, t) indicates that there is a relationship r between the head node h and the tail node t, since there are different products: the association of similar types, substitution, combination and the like with different strengths needs to be specific to a specific scene. For example, in an IPTV home online tv scenario, for movie resources, series a is similar to series B, which is associated with series C (the leading corners of B and C are the same actor).
Second, the user's interaction record with the project is modeled as a bipartite graph G 2 ＝{(u,act,i)|u∈U,i∈I 2 ,act∈R 2 }, using
Figure BDA0003662626020000084
And
Figure BDA0003662626020000085
respectively representing users and item sets, where M and N respectively represent the number of users and the number of items, and the interaction relationship between a user and an item is represented by a tuple τ (u, act, i), for example: purchase, collection, comment, on-demand, etc.
And then fusing the user interaction graph and the item relation graph to obtain a unified knowledge graph G { (h, R, t) | h, t ∈ E, R ∈ R }, wherein the set E ═ U { [ U ] U [ U ] I [ ] 1 The set R ═ R 1 ∪R 2 。
The knowledge graph based path reasoning module:
the path reasoning module defines the path between the user arriving at the project on G as the sequence of the entity and the relation:
Figure BDA0003662626020000091
wherein e 1 ＝u,e L ＝p；(e l ,r l ,e l+1 ) Is the ith tuple on path p and L-1 represents the number of all tuples on path p.
The path inference module infers for a given path p k A handle p k Each entity in (a) is partitioned into a corresponding type and value (e.g., entity type is user, value is minuscule) and initialized into two vectors
Figure BDA0003662626020000092
d represents the dimension size. Path p of the same way k The path in (1) is embedded as a vector r of the same dimension l Thus, a relationship path p is obtained k Is characterized in that: [ e ] a 1 ,r 1 ,e 2 ,...,e l-1 ,r l-1 ]Each element represents an entity or a type of relationship.
After a description path of an embedded sequence is obtained, sequence information can be mined by utilizing a recurrent neural network model to generate a single representation to encode the whole semantics of the sequence information, the LSTM is adopted in the invention, and at the l-1 path, the LSTM layer is input as a splicing vector x of three vectors of entity types, entity values and relations on the path l-1 The calculation method is as follows:
Figure BDA0003662626020000093
outputting a hidden state vector h l-1 The calculation method is as follows:
z l ＝tanh(W 1 x l +W h h l-1 +b z ) (2)
f l ＝σ(W f x l +W h h l-1 +b f ) (3)
i l ＝σ(W i x l +W h h l-1 +b i ) (4)
o l ＝σ(W o x l +W h h l-1 +b o ) (5)
c l ＝f l ⊙c l-1 +i l ⊙z l (6)
h l ＝o l ⊙tanh(c l ) (7)
in the formula c l ∈R d ,z∈R d Respectively representing the unit storage state vector and the information conversion module, d' representing the number of hidden units, i l ,o l And f l Respectively representing input, output and forgetting gate, W z ,W i ,W f ,W o ∈R d′×3d ,W h ∈R d′×d′ Representing a mapping coefficient matrix; and b z ,b i ,b f And W o Is an offset vector, σ () is an activation function, which indicates the element-level product between two vectors. Finally outputting the state vector h by using a memory mechanism L The whole path p can be represented K 。
A score prediction module:
the score prediction module is for a given user u and a target item i, and u can follow a series of path sets P (u, i) ═ P to reach i in G 1 ,p 2 ,...,p K }; outputting a prediction score of u and i interaction:
Figure BDA0003662626020000101
in the formula (I), the compound is shown in the specification,
Figure BDA0003662626020000102
to evaluate the predicted outcome; f. of θ A model with a parameter theta is represented, which indicates whether the user u interacts with the item i is inferred from the connectivity p (u, i).
And the generation recommendation module projects the final state into the output prediction score by adopting two fully-connected layers. The prediction formula is:
Figure BDA0003662626020000103
w1, W2 denote coefficient weights for layer 1 and layer 2, respectively, and Relu denotes an activation function. Where τ ═ (u, act, i) represents a tuple; s (τ | p) k )＝s k Representing the predicted score of the k path from the node u to the node i, and b representing the offset;
the generate recommendation module designs a weighted pooling operation to aggregate the predicted scores of the interaction tuples:
Figure BDA0003662626020000104
wherein
Figure BDA0003662626020000105
Where γ is a hyperparameter controlling the weight of each index, { s 1 ,s 2 ,...,s K Expressing the respective prediction scores of K paths from the node u to the node i; such pooling enables the differentiation of different path importance.

Claims (10)

1. A multi-user recommendation system based on knowledge graph path inference is characterized in that: the system comprises a knowledge graph construction module, the path reasoning module and a score prediction module.
The knowledge graph construction module acquires user interaction historical data, constructs a knowledge graph G and transmits the knowledge graph G to the path reasoning module.
The path reasoning module generates a plurality of relation paths between the users and the target project according to the knowledge graph G and transmits the relation paths to the grading prediction module;
and the scoring prediction module carries out evaluation prediction on the received relation paths and outputs a project recommendation list to the user according to the evaluation prediction result.
2. The multi-user recommendation system based on knowledge-graph path inference as claimed in claim 1, wherein the step of constructing a knowledge-graph G comprises:
1) building item associated directed graph G 1 (ii) a The item association directed graph G 1 The entity of (1) comprises an interactive item and a candidate item, and the edge set comprises an incidence relation between the entity and the entity;
2) building a directed graph G of user-item interactions 2 (ii) a The interaction directed graph G 2 The entities of (1) include users and items; the interaction directed graph G 2 If the user and the project have interaction, an edge exists between the user and the project;
3) associating items with directed graph G 1 And user interaction directed graph G with items 2 And fusing to obtain a knowledge graph G.
3. The multi-user recommendation system based on knowledge-graph path inference as claimed in claim 2, wherein the item-associated directed graph G 1 ＝{(h,r,t)|h,t∈I 1 ,r∈R 1 Obtaining through modeling of associated data between the candidate items and the interacted items;
wherein the tuple (h, r, t) indicates that a relationship r exists between the head node h and the tail node t; i is 1 All candidate item sets are selected; r 1 Is a set of relationships.
4. The multi-user recommendation system based on knowledge-graph path inference as claimed in claim 3, characterized in that the interaction directed graph G of users and items 2 ＝{(u,act,i)|u∈U,i∈I 2 ,act∈R 2 The method is obtained by modeling of interaction records of users and projects;
wherein, the tuple (u, act, i) represents that the interaction relation act exists between the user u and the item i;
Figure FDA0003662626010000011
and
Figure FDA0003662626010000012
respectively representing users, item sets, M and N respectively representing the number of users and the number of items,
Figure FDA0003662626010000013
R 2 a set of interactions is represented.
5. The system of claim 4, wherein the knowledgegraph G { (h, R, t) | h, t ∈ E, R ∈ R }, wherein the set E { [ U ] I [ } 1 The set R ═ R 1 ∪R 2 。
6. The multi-user recommendation system based on knowledge-graph path inference according to claim 1, wherein the step of generating relationship paths between users and target items by the path inference module comprises:
1) recording the path between the user and the project on the knowledge graph G as the sequence of entities and relations
Figure FDA0003662626010000021
Wherein e is 1 ＝u,e L ＝p；(e l ,r l ,e l+1 ) Is the ith tuple on path p, L-1 represents the number of all tuples on path p;
2) for a given relationship path p on the knowledge-graph G k Said path inference module links the relationship path p k Each entity in (1) is partitioned into a corresponding type and value, and the type and value are initialized into two vectors which are respectively recorded as
Figure FDA0003662626010000022
d represents the dimension size; k is initially 1;
3) on the path p k In which vectors r of the same dimension are embedded l Updating relationship path p' k Is [ e ] 1 ,r 1 ,e 2 ,...,e l-1 ,r l-1 ]Relation path p' k Each element in the list represents an entity or a type of relationship;
4) mining relationship path p 'using a recurrent neural network model stored on a path inference module' k Thereby generating an encoded path p k ；
5) Let k be k +1 and return to step 2) until the encoded path set P (u, i) is obtained as { P } 1 ,p 2 ,...,p K }。
7. The system of claim 6, wherein the recurrent neural network model is an LSTM neural network model, comprising an input layer, a hidden layer, and an output layer; the hidden layer comprises an input gate, a forgetting gate and an output gate;
the input of the recurrent neural network model comprises entity type, entity value and relation vector, and the output is the encoded path.
8. The system of claim 7, wherein the input vector x of the recurrent neural network model is l-1 Output vector h L Respectively satisfy the following formula:
Figure FDA0003662626010000023
z l ＝tanh(W 1 x l +W h h l-1 +b z ) (2)
f l ＝σ(W f x l +W h h l-1 +b f ) (3)
i l ＝σ(W i x l +W h h l-1 +b i ) (4)
o l ＝σ(W o x l +W h h l-1 +b o ) (5)
c l ＝f l ⊙c l-1 +i l ⊙z l (6)
h l ＝o l ⊙tanh(c l ) (7)
in the formula, c l ∈R d ,z l ∈R d Respectively representing the unit storage state vector and the information conversion module, and d' represents the number of the hidden units; i.e. i l 、o l And f l Respectively representing input, output and forgetting gate, W z 、W i 、W f 、W o ∈R d'×3d ,W h ∈R d'×d' Representing a mapping coefficient matrix; b z 、b i 、b f 、b o Is a bias vector; σ (.) is an activation function, an indication of an element-level product between two vectors; e.g. of the type l-1 、e′ l-1 、r l-1 Respectively representing entity types, entity values and relationship vectors.
9. The system of claim 1, wherein the scoring and predicting module evaluates and predicts the received relationship paths by:
1) establishing a prediction function, namely:
Figure FDA0003662626010000031
in the formula (I), the compound is shown in the specification,
Figure FDA0003662626010000032
to evaluate the predicted outcome; f. of θ Representing a model with a parameter θ, equation (8) indicates whether user u interacts with item i is determined by connectivity ρ (u, i);
2) calculating a state prediction value, namely:
Figure FDA0003662626010000033
in the formula, W 1 ，W 2 Respectively representing coefficient weights of a 1 st layer full connection layer and a 2 nd layer full connection layer; relu denotes an activation function; where τ ═ (u, act, i) denotes a tuple; s (τ | p) k )＝s k Representing the predicted score of the k path from the node u to the node i, and b representing the offset;
3) projecting the state predicted value into a prediction function to obtain an evaluation prediction result
Figure FDA0003662626010000034
Namely:
Figure FDA0003662626010000035
where σ () is the activation function, where the parameter g(s) 1 ,s 2 ,...,s K ) As follows:
Figure FDA0003662626010000036
where γ is a hyperparameter controlling each exponential weight; { s 1 ,s 2 ,...,s K Expressing the respective prediction scores of K paths from the node u to the node i; s is k And representing the predicted score of the k-th path from the node u to the node i.
10. The multi-user recommendation system based on knowledge-graph path inference as claimed in claim 1, wherein: the items in the item recommendation list are predicted results according to scores
Figure FDA0003662626010000041
The top N items in the descending order.
CN202210583484.1A 2022-05-25 2022-05-25 Multi-user recommendation system based on knowledge graph path reasoning Active CN114820139B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210583484.1A CN114820139B (en) 2022-05-25 2022-05-25 Multi-user recommendation system based on knowledge graph path reasoning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210583484.1A CN114820139B (en) 2022-05-25 2022-05-25 Multi-user recommendation system based on knowledge graph path reasoning

Publications (2)

Publication Number Publication Date
CN114820139A true CN114820139A (en) 2022-07-29
CN114820139B CN114820139B (en) 2024-05-28

Family

ID=82519194

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210583484.1A Active CN114820139B (en) 2022-05-25 2022-05-25 Multi-user recommendation system based on knowledge graph path reasoning

Country Status (1)

Country Link
CN (1) CN114820139B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115618108A (en) * 2022-10-24 2023-01-17 云南中烟工业有限责任公司 Cigarette brand recommendation method based on knowledge graph in new retail mode
CN116049438A (en) * 2023-01-10 2023-05-02 昆明理工大学 Knowledge graph-based group membership analysis method

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112417306A (en) * 2020-12-10 2021-02-26 北京工业大学 Method for optimizing performance of recommendation algorithm based on knowledge graph
CN112884548A (en) * 2021-02-01 2021-06-01 北京三快在线科技有限公司 Object recommendation method and device based on path reasoning and electronic equipment
CN113326384A (en) * 2021-06-22 2021-08-31 四川大学 Construction method of interpretable recommendation model based on knowledge graph
WO2021189971A1 (en) * 2020-10-26 2021-09-30 平安科技（深圳）有限公司 Medical plan recommendation system and method based on knowledge graph representation learning
US20210406779A1 (en) * 2020-06-26 2021-12-30 Facebook, Inc. System and methods for feature engineering based on graph learning
US20220058489A1 (en) * 2020-08-19 2022-02-24 The Toronto-Dominion Bank Two-headed attention fused autoencoder for context-aware recommendation
CN114139066A (en) * 2021-09-10 2022-03-04 重庆大学 Collaborative filtering recommendation system based on graph neural network

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210406779A1 (en) * 2020-06-26 2021-12-30 Facebook, Inc. System and methods for feature engineering based on graph learning
US20220058489A1 (en) * 2020-08-19 2022-02-24 The Toronto-Dominion Bank Two-headed attention fused autoencoder for context-aware recommendation
WO2021189971A1 (en) * 2020-10-26 2021-09-30 平安科技（深圳）有限公司 Medical plan recommendation system and method based on knowledge graph representation learning
CN112417306A (en) * 2020-12-10 2021-02-26 北京工业大学 Method for optimizing performance of recommendation algorithm based on knowledge graph
CN112884548A (en) * 2021-02-01 2021-06-01 北京三快在线科技有限公司 Object recommendation method and device based on path reasoning and electronic equipment
CN113326384A (en) * 2021-06-22 2021-08-31 四川大学 Construction method of interpretable recommendation model based on knowledge graph
CN114139066A (en) * 2021-09-10 2022-03-04 重庆大学 Collaborative filtering recommendation system based on graph neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
王钰蓥: "基于知识图谱和Bi-LSTM的推荐算法", 《计算机与现代化》, no. 09, 15 September 2021 (2021-09-15), pages 90 - 98 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115618108A (en) * 2022-10-24 2023-01-17 云南中烟工业有限责任公司 Cigarette brand recommendation method based on knowledge graph in new retail mode
CN115618108B (en) * 2022-10-24 2023-05-26 云南中烟工业有限责任公司 Brand recommendation method based on knowledge graph in new retail model
CN116049438A (en) * 2023-01-10 2023-05-02 昆明理工大学 Knowledge graph-based group membership analysis method
CN116049438B (en) * 2023-01-10 2023-06-02 昆明理工大学 Knowledge graph-based group membership analysis method

Also Published As

Publication number Publication date
CN114820139B (en) 2024-05-28

Similar Documents

Publication Publication Date Title
CN110837598B (en) Information recommendation method, device, equipment and storage medium
CN109190030B (en) Implicit feedback recommendation method fusing node2vec and deep neural network
CN114820139B (en) Multi-user recommendation system based on knowledge graph path reasoning
CN110910218A (en) Multi-behavior migration recommendation method based on deep learning
CN111881350A (en) Recommendation method and system based on mixed graph structured modeling
CN115618108B (en) Brand recommendation method based on knowledge graph in new retail model
CN115270001B (en) Privacy protection recommendation method and system based on cloud collaborative learning
CN112819575B (en) Session recommendation method considering repeated purchasing behavior
Abolghasemi et al. A personality-aware group recommendation system based on pairwise preferences
CN114386513A (en) Interactive grading prediction method and system integrating comment and grading
Ahamed et al. A recommender system based on deep neural network and matrix factorization for collaborative filtering
CN112685657A (en) Conversation social recommendation method based on multi-mode cross fusion graph network
CN115618128A (en) Collaborative filtering recommendation system and method based on graph attention neural network
CN112559904B (en) Conversational social recommendation method based on door mechanism and multi-modal graph network
Zhang et al. Knowledge graph driven recommendation model of graph neural network
CN112364245B (en) Top-K movie recommendation method based on heterogeneous information network embedding
CN113761388A (en) Recommendation method and device, electronic equipment and storage medium
CN114139066A (en) Collaborative filtering recommendation system based on graph neural network
CN115809339A (en) Cross-domain recommendation method, system, device and storage medium
CN116484092A (en) Hierarchical attention network sequence recommendation method based on long-short-term preference of user
Sangeetha et al. Predicting personalized recommendations using GNN
CN115391555A (en) User-perceived knowledge map recommendation system and method
CN113849725A (en) Socialized recommendation method and system based on graph attention confrontation network
Thirunavukarasu et al. People to people recommendation using coupled nonnegative Boolean matrix factorization
CN113704440B (en) Conversation recommendation method based on path representation in article graph network

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