CN113781150A - Article recommendation method and device - Google Patents

Article recommendation method and device Download PDF

Info

Publication number
CN113781150A
CN113781150A CN202110119534.6A CN202110119534A CN113781150A CN 113781150 A CN113781150 A CN 113781150A CN 202110119534 A CN202110119534 A CN 202110119534A CN 113781150 A CN113781150 A CN 113781150A
Authority
CN
China
Prior art keywords
article
user
feature vector
vector
item
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
CN202110119534.6A
Other languages
Chinese (zh)
Other versions
CN113781150B (en
Inventor
陈玉杰
付岐峰
郑冬
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.)
Beijing Jingdong Century Trading Co Ltd
Beijing Wodong Tianjun Information Technology Co Ltd
Original Assignee
Beijing Jingdong Century Trading Co Ltd
Beijing Wodong Tianjun Information Technology Co Ltd
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 Beijing Jingdong Century Trading Co Ltd, Beijing Wodong Tianjun Information Technology Co Ltd filed Critical Beijing Jingdong Century Trading Co Ltd
Priority to CN202110119534.6A priority Critical patent/CN113781150B/en
Publication of CN113781150A publication Critical patent/CN113781150A/en
Application granted granted Critical
Publication of CN113781150B publication Critical patent/CN113781150B/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
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses an article recommendation method and device, and relates to the technical field of computers. One embodiment of the method comprises: generating a first feature vector of the article according to the article information; generating a first feature vector of a user according to the user information; determining a user relationship network and an article relationship network; training a scoring model according to the first feature vector of the item, the first feature vector of the user, the user relationship network and the item relationship network; and recommending the articles to the user according to the trained scoring model. This embodiment can improve recommendation diversity.

Description

Article recommendation method and device
Technical Field
The invention relates to the technical field of computers, in particular to an article recommendation method and device.
Background
In order to provide more satisfactory articles for the user, the e-commerce platform extracts user characteristics based on browsing records, clicking behaviors and the like of the user and provides personalized recommendation for the user according to the extracted characteristics.
Existing recommendation methods generally recommend items to a user based only on the relationship between the user's behavioral data and the user.
The recommended articles are single through the method.
Disclosure of Invention
In view of this, embodiments of the present invention provide an article recommendation method and apparatus, which can improve recommendation diversity.
In a first aspect, an embodiment of the present invention provides an item recommendation method, including:
generating a first feature vector of the article according to the article information;
generating a first feature vector of a user according to the user information;
determining a user relationship network and an article relationship network;
training a scoring model according to the first feature vector of the article, the first feature vector of the user, the user relationship network and the article relationship network;
and recommending the articles to the user according to the trained scoring model.
Alternatively,
the generating a first feature vector of the article according to the article information includes:
respectively extracting article characteristics from the multi-modal article information;
and splicing the article features of the article information of each mode to obtain a first feature vector of the article.
Alternatively,
the modalities include: a text;
the extracting of the article features from the multimodal article information, respectively, comprises:
inputting the article information into word2vec to obtain a word vector;
and inputting the word vector into a trained first convolutional neural network to obtain the article characteristics.
Alternatively,
the modalities include: a picture;
the extracting of the article features from the multimodal article information, respectively, comprises:
and inputting the article information into a trained second convolutional neural network to obtain the article characteristics.
Alternatively,
the user information comprises: attribute information and a browsing item sequence;
the generating a first feature vector of the user according to the user information includes:
inputting the attribute information into an embedding layer to obtain a second feature vector of the user;
and inputting the second characteristic vector of the user and the first characteristic vector of each browsed article in the browsed article sequence into a trained vector model to obtain the first characteristic vector of the user.
Alternatively,
the inputting the second feature vector of the user and the first feature vector of each browsed article in the browsed article sequence into a trained vector model to obtain the first feature vector of the user includes:
respectively inputting the first feature vector of each browsed article into a gate control circulation unit to obtain a context vector of each browsed article;
inputting the context vector of each of the viewed items and the second feature vector of the user into an attention layer, so that the attention layer calculates a weight of each of the viewed items according to the context vector of each of the viewed items and the second feature vector of the user; determining a first feature vector of the user based on the context vector and the weight of each of the viewed items.
Alternatively,
the determining an item relationship network comprises:
calculating the similarity of the two articles according to the positions of the articles;
generating the article relation network by taking articles as nodes and the similarity of the two articles as edges; wherein, in the item relationship network, if the similarity of two items is larger than a specified threshold, an edge exists between the items.
Alternatively,
the training a scoring model according to the first feature vector of the item, the first feature vector of the user, the user relationship network, and the item relationship network includes:
generating a user adjacency matrix according to the user relationship network;
generating an article adjacency matrix according to the article relation network;
inputting the first eigenvector of the user and the user adjacency matrix into a first graph neural network to obtain a third eigenvector of the user;
inputting the first feature vector of the article and the article adjacency matrix into a second graph neural network to obtain a second feature vector of the article;
inputting the third feature vector of the user and the second feature vector of the article into a multilayer perceptron to obtain a prediction score of the article;
and adjusting parameters of the scoring model according to the prediction score of the article and a preset loss function.
In a second aspect, an embodiment of the present invention provides an article recommendation apparatus, including:
the generating module is configured to generate a first feature vector of the article according to the article information; generating a first feature vector of a user according to the user information;
a determination module configured to determine a user relationship network and an item relationship network;
a prediction module configured to train a scoring model based on the first feature vector of the item, the first feature vector of the user, the user relationship network, and the item relationship network;
and the recommending module is configured to recommend the articles to the user according to the trained scoring model.
In a third aspect, an embodiment of the present invention provides an electronic device, including:
one or more processors;
a storage device for storing one or more programs,
when executed by the one or more processors, cause the one or more processors to implement the method of any of the embodiments described above.
In a fourth aspect, an embodiment of the present invention provides a computer-readable medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the method of any one of the above embodiments.
One embodiment of the above invention has the following advantages or benefits: the scoring model is trained through the first feature vector of the article, the first feature vector of the user, the user relation network and the article relation network, so that the scoring model not only learns the self features of the user and the self features of the article, but also learns the association relationship among the users and the association relationship among the articles, and the recommendation diversity can be improved.
Further effects of the above-mentioned non-conventional alternatives will be described below in connection with the embodiments.
Drawings
The drawings are included to provide a better understanding of the invention and are not to be construed as unduly limiting the invention. Wherein:
FIG. 1 is a flow chart of a method for recommending items according to an embodiment of the present invention;
FIG. 2 is a flow chart of a method for recommending items according to another embodiment of the present invention;
FIG. 3 is a schematic diagram of an item recommendation device provided in accordance with an embodiment of the present invention;
FIG. 4 is an exemplary system architecture diagram in which embodiments of the present invention may be employed;
fig. 5 is a schematic block diagram of a computer system suitable for use in implementing a terminal device or server of an embodiment of the invention.
Detailed Description
Exemplary embodiments of the present invention are described below with reference to the accompanying drawings, in which various details of embodiments of the invention are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the invention. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
As shown in fig. 1, an embodiment of the present invention provides an item recommendation method, including:
step 101: according to the item information, a first feature vector of the item is generated.
The article information can be in various modes such as characters, pictures, audio and video. For example, there may be a piece of text describing the item, and there may also be a picture including the item. A first feature vector of the article is extracted from the article information.
Step 102: and generating a first feature vector of the user according to the user information.
The user information may include: attribute information, behavior information, etc. The attribute information includes: the age, sex, and occupation of the user. The behavior information includes: user browsing through a sequence of items, clicking on a record of items, etc.
Step 103: a user relationship network and an item relationship network are determined.
The user relationship network takes users as nodes and takes social relationships among the users as edges. The social relationship can be determined according to a mobile phone address book, a social software address book and the like of the user. The article relation network takes articles as nodes and similarity among the articles as edges. For example, if both items belong to the fruit category, an edge exists between the two items.
Step 104: and training a scoring model according to the first feature vector of the article, the first feature vector of the user, the user relationship network and the article relationship network.
The scoring model may include: the system comprises a graph volume network and a multi-layer perceptron, wherein the multi-layer perceptron is composed of a full connection layer and a softmax layer. The training process may use cross entropy as a loss function and may also use least squares difference as a loss function.
Step 105: and recommending the articles to the user according to the trained scoring model.
The scoring of the user on each item can be obtained through the scoring model, and the items recommended to the user can be selected according to the order of the scores from high to low.
The scoring model is trained through the first feature vector of the article, the first feature vector of the user, the user relation network and the article relation network, so that the scoring model not only learns the self features of the user and the self features of the article, but also learns the association relationship among the users and the association relationship among the articles, and the recommendation diversity can be improved.
In one embodiment of the present invention, generating a first feature vector of an item according to item information includes:
respectively extracting article characteristics from the multi-modal article information;
and splicing the article features of the article information of each mode to obtain a first feature vector of the article.
In an actual application scenario, the article information may have only one modality, or may have multiple modalities, and if there are multiple modalities, article features corresponding to the modalities need to be spliced. By splicing the multi-modal article characteristics, the article can be more accurately characterized so as to obtain a more accurate recommendation result.
In one embodiment of the present invention, if the item information is text, extracting item features from the multimodal item information respectively comprises:
inputting article information into word2vec to obtain a word vector;
and inputting the word vector into the trained first convolution neural network to obtain the article characteristics.
Mapping each word in the text into a vector with a specified dimension by using word2vec, and then inputting the vector into a CNN (Convolutional Neural Networks) for feature extraction.
In an embodiment of the present invention, if the article information is a picture, extracting article features from the multimodal article information respectively includes:
and inputting the article information into the trained second convolutional neural network to obtain article characteristics.
In the embodiment of the present invention, the article features are extracted from the picture through the CNN, and other neural networks may also be used to perform feature extraction, for example, a Histogram of Oriented Gradients (HOG).
In one embodiment of the present invention, the user information includes: attribute information and a browsing item sequence;
generating a first feature vector of a user according to the user information, wherein the first feature vector comprises:
inputting the attribute information into the embedded layer to obtain a second feature vector of the user;
and inputting the second characteristic vector of the user and the first characteristic vector of each browsed article in the browsed article sequence into the trained vector model to obtain the first characteristic vector of the user.
The attribute information includes: occupation, age, gender, and user id. The Embedding Layer (Embedding Layer) is used for digitizing the attribute information. The attribute information can reflect a long-term interest of the user, and the browsing item sequence can reflect a short-term interest of the user. And the first feature vector of the user is determined by the attribute information and the browsing item sequence, and can reflect the long-term interest and the short-term interest of the user. Long-term interest refers to a user remaining interested in an item for a longer period of time, e.g., a lawyer may have long-term interest in a legal book, while short-term interest refers to a user interested in an item for a shorter period of time, e.g., if a trash can in a home is bad, the user may browse the trash can's information within a few days and no longer be concerned after buying it.
In an actual application scenario, the short-term interest of the user can be mined from the collected item sequence and the searched item sequence, that is, the browsed item sequence is replaced by the collected item sequence or the searched item sequence, and the browsed item sequence and the collected item sequence can be combined.
The first feature vector of the user fuses long-term interest and short-term interest of the user, and recommendation accuracy can be improved.
In an embodiment of the present invention, inputting the second feature vector of the user and the first feature vector of each viewed item in the sequence of viewed items into the trained vector model to obtain the first feature vector of the user, includes:
respectively inputting the first feature vector of each browsed article into a gate control circulation unit to obtain a context vector of each browsed article;
inputting the context vector of each browsed item and the second feature vector of the user into an attention layer, so that the attention layer calculates the weight of each browsed item according to the context vector of each browsed item and the second feature vector of the user; a first feature vector of the user is determined based on the context vector and the weights of the respective viewed items.
In an embodiment of the present invention, the vector model includes a GRU (Gate recovery Unit) and an attention layer. In order to distinguish the training and prediction processes of the vector model, the user participating in the training is a first user, and the user participating in the prediction process is a second user, in the embodiment of the invention, the parameters of the vector model are adjusted through the second feature vector of the first user and the corresponding first feature vectors of all browsed articles, and the second feature vector of the second user and the corresponding first feature vectors of all browsed articles are input into the trained vector model to obtain the first feature vector of the second user.
The preference degrees of the users are different for each browsed item, and the preference degrees of the users for each item are determined through the attention mechanism. Wherein the weight of the browsing item is used for representing the preference degree of the user for the browsing item.
Considering that similar items tend to aggregate with each other, e.g., clothing stores tend to be set up in shopping malls and breakfast shops tend to be set up around residences, embodiments of the present invention establish an item relationship network based on the location of items.
Specifically, determining an item relationship network comprises:
calculating the similarity of the two articles according to the positions of the articles;
taking an article as a node and the similarity of two articles as an edge to generate an article relation network; in the article relation network, if the similarity of two articles is larger than a specified threshold, an edge exists between the articles.
According to the embodiment of the invention, the Euclidean distance between the articles can be calculated according to the positions of the articles, and the Euclidean distance between the articles is taken as the similarity of the two articles. The larger the euclidean distance, the lower the similarity. In the embodiment of the invention, in order to improve the recommendation efficiency, an edge may be established between the two items when the similarity of the two items is greater than a specified threshold.
In an actual application scenario, the similarity between the articles can be determined based on the name, type, manufacturer and other information of the articles.
Training a scoring model according to the first feature vector of the item, the first feature vector of the user, the user relationship network and the item relationship network, wherein the training comprises the following steps:
generating a user adjacency matrix according to the user relationship network;
generating an article adjacency matrix according to the article relation network;
inputting the first eigenvector of the user and the user adjacency matrix into the first graph neural network to obtain a third eigenvector of the user;
inputting the first feature vector of the article and the article adjacency matrix into a second graph neural network to obtain a second feature vector of the article;
inputting the third feature vector of the user and the second feature vector of the article into a multilayer perceptron to obtain a prediction score of the article;
and adjusting parameters of the scoring model according to the prediction scoring of the article and a preset loss function.
In the embodiment of the present invention, the user adjacency matrix is used to represent social relationships between users, i.e., edges of the user relationship network, and the article adjacency matrix is used to represent similarities between articles, i.e., edges of the article relationship network, both of which are two-dimensional arrays.
According to the embodiment of the invention, the preference of the similar user can be learned through the first graph neural network, the articles preferred by the associated user can be recommended to the user, and the articles similar to the articles browsed by the user can be recommended to the user through the characteristics of the similar articles learned through the second graph neural network. The first graph neural network and the second graph neural network are combined, so that the diversity of item recommendation can be improved. As shown in fig. 2, an embodiment of the present invention provides an item recommendation method, including:
step 201: and inputting the article text into word2vec to obtain a word vector, and inputting the word vector into the trained first convolution neural network to obtain article text characteristics.
The embodiment of the invention acquires the information of 5000 users and the information of 5000 articles from the data set Yelp 2018.
And mapping each word in the article text into a 200-dimensional vector by using word2vec, inputting the 200-dimensional vector into the CNN, and performing feature extraction on the word vector to obtain article text features.
Step 202: and inputting the article picture into the trained second convolutional neural network to obtain the article picture characteristic.
And inputting the article picture into the CNN to obtain the article picture characteristic.
Step 203: and splicing the article text characteristic and the article picture characteristic to obtain a first characteristic vector of the article.
Step 204: and inputting the attribute information of the user into the embedding layer to obtain a second feature vector of the user.
And inputting the occupation, age, gender and user id of the user into the embedding layer to obtain a second feature vector of the user.
Step 205: and respectively inputting the first characteristic vector of each browsed article in the sequence of the browsed articles of the user into a gate control circulation unit to obtain the context vector of each browsed article.
Browsing item sequences as p1To pk。
Extracting context vectors for browsing items through the GRU:
Figure BDA0002921947810000091
wherein p iskParameters for characterizing viewed items k, theta for characterizing GRUs,
Figure BDA0002921947810000092
a first feature vector for characterizing an item k,
Figure BDA0002921947810000096
for characterizing the context vector for browsing item k. The hidden layer unit dimension of the GRU is 100 and the output vector dimension is 200.
In an embodiment of the invention, in order to improve the training speed, the number of browsing items in the browsing item sequence is less than 30.
Step 206: inputting the context vector of each browsed item and the second feature vector of the user into an attention layer, so that the attention layer calculates the weight of each browsed item according to the context vector of each browsed item and the second feature vector of the user; a first feature vector of the user is determined based on the context vector and the weights of the respective viewed items.
And (3) calculating the weight of the browsed goods.
Figure BDA0002921947810000093
Figure BDA0002921947810000094
Wherein alpha iskFor characterizing the weight of the viewed item k, v for characterizing the weight matrix, WHAs context vector weights, WXFor characterizing the first feature vector weight, b is a constant,
Figure BDA0002921947810000095
a first feature vector for characterizing user i.
A first feature vector of the user is calculated by equation (4).
Figure BDA0002921947810000101
The embodiment of the invention takes the second feature vector of the user as the initialization state of the first feature vector of the user, uses the hidden state of the GRU network to represent the context vector of the browsed articles, and finally generates the first feature vector of the user which combines long-term interest and short-term interest.
Step 207: and calculating the similarity of the two articles according to the positions of the articles, and generating an article relation network by taking the articles as nodes and the similarity of the two articles as edges.
In the article relation network, if the similarity of two articles is larger than a specified threshold, an edge exists between the articles.
The location of the item may be expressed in terms of the latitude and longitude in which the item is located. The specified threshold may be 30 kilometers, with an edge between two items if their euclidean distance is below 30 kilometers.
Step 208: and determining the user relationship network and the article relationship network by taking the users as nodes and taking the social relationship among the users as edges.
Step 209: and generating a user adjacency matrix according to the user relation network.
Step 210: and generating an article adjacency matrix according to the article relation network.
Step 211: and inputting the first feature vector of the user and the user adjacency matrix into the first graph neural network to obtain a third feature vector of the user.
In the embodiment of the invention, the number of users is M, and the first characteristic matrix X of the usersUMay be represented by formula (5).
Figure BDA0002921947810000102
The third eigenvector matrix U for the user is represented by equation (6).
U＝GCN(2)(XU；AU，θG，U) (6)
Wherein A isUFor the user adjacency matrix, θG，UParameters used to characterize the GCN. In order to ensure the recommendation accuracy and improve the training speed, the embodiment of the invention adopts two layers of convolution networks.
Since the farther two nodes are away from each other in the social network, the weaker the information association strength between the two nodes is, the smaller the influence is. Therefore, the embodiment of the invention can only pay attention to the user information in two nodes away from each other, thereby improving the training efficiency. For example, user 2 is connected to user 1, user 3 is connected to user 2 vector, user 4 is connected to user 3, and if the third feature vector of user 1 is to be determined, only user 2 and user 3 are concerned, and user 4 is not concerned.
Step 212: and inputting the first feature vector of the article and the article adjacency matrix into a second graph neural network to obtain a second feature vector of the article.
In the embodiment of the invention, the quantity of the articles is N, and the first feature matrix Y of the articlesPCan be expressed by the formula (7).
Figure BDA0002921947810000111
The second eigenvector matrix P for the article is represented by equation (8).
P＝GCN(2)(XP；AP，θG，P) (8)
Wherein A isPIs an article abutment matrix, thetaG，PParameters used to characterize the GCN. In order to ensure the recommendation accuracy and the recommendation efficiency, the embodiment of the invention adopts two layers of convolution networks.
Step 213: and inputting the third feature vector of the user and the second feature vector of the article into a multilayer perceptron to obtain a prediction score of the article.
Inputting the third feature vector of the user and the second feature vector of the article into a multi-layer feelingKnowing the machine to obtain the prediction score of the user u on the item p
Figure BDA0002921947810000112
As shown in equation (9).
Figure BDA0002921947810000113
Wherein f ismlp(. for characterizing a multi-layer perceptron, θ)rParameters for characterizing a multi-layer perceptron.
Step 214: and adjusting parameters of the scoring model according to the prediction scoring of the article and a preset loss function.
The loss function employed in the present invention is equation (10).
Figure BDA0002921947810000114
Wherein,
Figure BDA0002921947810000115
for characterizing the prediction score, rupFor characterizing the truth score.
Step 215: and recommending the articles to the user according to the trained scoring model.
In the prediction process, the scoring model can determine the prediction scores of the target users for all the articles, and according to the sequence of the prediction scores from high to low, a specified number of articles are selected and recommended to the target users.
The embodiment of the invention is based on the Yelp 2018 dataset, and different recommendation methods are evaluated through Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), and the results are shown in Table 1.
The recommendation method comprises the following steps: improved singular value decomposition (SVD + +), matrix decomposition-Bayesian personalized ranking (MF-BPR), dual interest model based (D-ATT), deep collaborative neural network (DeepCoNN), neural attention score regression for interpretation of comment levels (NARRE), multi-pointer mutual attention network (MPCN).
As can be seen from table 1, the recommendation method provided by the embodiment of the present invention is superior to other recommendation methods.
TABLE 1 evaluation results of different recommendation methods
Recommendation method RMSE MAE
SVD++ 1.0968 0.7798
MF-BPR 1.0154 0.7723
D-ATT 0.9833 0.7599
DeepCoNN 0.8971 0.7562
NARRE 0.8875 0.7493
MPCN 0.8526 0.7285
The inventionExamples provide methods 0.7918 0.7043
The embodiment of the invention integrates the long-term interest and the short-term interest of the users, considers the correlation among the users and the correlation among the articles, and can improve the recommendation accuracy and diversity.
As shown in fig. 3, an embodiment of the present invention provides an article recommendation apparatus, including:
a generating module 301 configured to generate a first feature vector of an item according to the item information; generating a first feature vector of a user according to the user information;
a determination module 302 configured to determine a user relationship network and an item relationship network;
the prediction module 303 is configured to train a scoring model according to the first feature vector of the item, the first feature vector of the user, the user relationship network, and the item relationship network;
and the recommending module 304 is configured to recommend the articles to the user according to the trained scoring model.
In one embodiment of the present invention, the generating module 301 is configured to extract the item features from the multimodal item information, respectively; and splicing the article features of the article information of each mode to obtain a first feature vector of the article.
In one embodiment of the invention, the modalities include: a text; the generating module 301 is configured to input article information into word2vec to obtain a word vector; and inputting the word vector into the trained first convolution neural network to obtain the article characteristics.
In one embodiment of the invention, the modalities include: a picture; the generating module 301 is configured to input the article information into the trained second convolutional neural network to obtain the article features.
In one embodiment of the present invention, the user information includes: attribute information and a browsing item sequence; the generating module 301 is configured to input the attribute information into the embedding layer to obtain a second feature vector of the user; and inputting the second characteristic vector of the user and the first characteristic vector of each browsed article in the browsed article sequence into the trained vector model to obtain the first characteristic vector of the user.
In an embodiment of the present invention, the generating module 301 is configured to input the first feature vector of each browsed item into the gate control loop unit, respectively, to obtain a context vector of each browsed item; inputting the context vector of each browsed item and the second feature vector of the user into an attention layer, so that the attention layer calculates the weight of each browsed item according to the context vector of each browsed item and the second feature vector of the user; a first feature vector of the user is determined based on the context vector and the weights of the respective viewed items.
In one embodiment of the invention, the determining module 302 is configured to calculate the similarity between two items according to the positions of the items; taking an article as a node and the similarity of two articles as an edge to generate an article relation network; in the article relation network, if the similarity of two articles is larger than a specified threshold, an edge exists between the articles.
In an embodiment of the present invention, the prediction module 303 is configured to generate a user adjacency matrix according to a user relationship network; generating an article adjacency matrix according to the article relation network; inputting the first eigenvector of the user and the user adjacency matrix into the first graph neural network to obtain a third eigenvector of the user; inputting the first feature vector of the article and the article adjacency matrix into a second graph neural network to obtain a second feature vector of the article; inputting the third feature vector of the user and the second feature vector of the article into a multilayer perceptron to obtain a prediction score of the article; and adjusting parameters of the scoring model according to the prediction scoring of the article and a preset loss function.
An embodiment of the present invention provides an electronic device, including:
one or more processors;
a storage device for storing one or more programs,
when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the method of any of the embodiments as described above.
Embodiments of the present invention provide a computer-readable medium, on which a computer program is stored, which when executed by a processor implements the method according to any of the above embodiments.
Fig. 4 shows an exemplary system architecture 400 of an item recommendation method or an item recommendation apparatus to which embodiments of the present invention may be applied.
As shown in fig. 4, the system architecture 400 may include terminal devices 401, 402, 403, a network 404, and a server 405. The network 404 serves as a medium for providing communication links between the terminal devices 401, 402, 403 and the server 405. Network 404 may include various types of connections, such as wire, wireless communication links, or fiber optic cables, to name a few.
A user may use terminal devices 401, 402, 403 to interact with a server 405 over a network 404 to receive or send messages or the like. The terminal devices 401, 402, 403 may have installed thereon various communication client applications, such as shopping-like applications, web browser applications, search-like applications, instant messaging tools, mailbox clients, social platform software, etc. (by way of example only).
The terminal devices 401, 402, 403 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, desktop computers, and the like.
The server 405 may be a server providing various services, such as a background management server (for example only) providing support for shopping websites browsed by users using the terminal devices 401, 402, 403. The backend management server may analyze and perform other processing on the received data such as the product information query request, and feed back a processing result (for example, target push information, product information — just an example) to the terminal device.
It should be noted that the item recommendation method provided in the embodiment of the present invention is generally executed by the server 405, and accordingly, the item recommendation apparatus is generally disposed in the server 405.
It should be understood that the number of terminal devices, networks, and servers in fig. 4 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.
Referring now to FIG. 5, shown is a block diagram of a computer system 500 suitable for use with a terminal device implementing an embodiment of the present invention. The terminal device shown in fig. 5 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present invention.
As shown in fig. 5, the computer system 500 includes a Central Processing Unit (CPU)501 that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM)502 or a program loaded from a storage section 508 into a Random Access Memory (RAM) 503. In the RAM 503, various programs and data necessary for the operation of the system 500 are also stored. The CPU 501, ROM 502, and RAM 503 are connected to each other via a bus 504. An input/output (I/O) interface 505 is also connected to bus 504.
The following components are connected to the I/O interface 505: an input portion 506 including a keyboard, a mouse, and the like; an output portion 507 including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker; a storage portion 508 including a hard disk and the like; and a communication section 509 including a network interface card such as a LAN card, a modem, or the like. The communication section 509 performs communication processing via a network such as the internet. The driver 510 is also connected to the I/O interface 505 as necessary. A removable medium 511 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 510 as necessary, so that a computer program read out therefrom is mounted into the storage section 508 as necessary.
In particular, according to the embodiments of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method illustrated in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication section 509, and/or installed from the removable medium 511. The computer program performs the above-described functions defined in the system of the present invention when executed by the Central Processing Unit (CPU) 501.
It should be noted that the computer readable medium shown in the present invention can be a computer readable signal medium or a computer readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present invention, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present invention, however, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The modules described in the embodiments of the present invention may be implemented by software or hardware. The described modules may also be provided in a processor, which may be described as: a processor includes a sending module, an obtaining module, a determining module, and a first processing module. The names of these modules do not form a limitation on the modules themselves in some cases, and for example, the sending module may also be described as a "module sending a picture acquisition request to a connected server".
As another aspect, the present invention also provides a computer-readable medium that may be contained in the apparatus described in the above embodiments; or may be separate and not incorporated into the device. The computer readable medium carries one or more programs which, when executed by a device, cause the device to comprise:
generating a first feature vector of the article according to the article information;
generating a first feature vector of a user according to the user information;
determining a user relationship network and an article relationship network;
training a scoring model according to the first feature vector of the article, the first feature vector of the user, the user relationship network and the article relationship network;
and recommending the articles to the user according to the trained scoring model.
According to the technical scheme of the embodiment of the invention, the scoring model is trained through the first feature vector of the article, the first feature vector of the user, the user relationship network and the article relationship network, so that the scoring model not only learns the self features of the user and the self features of the article, but also learns the association relationship between the users and the association relationship between the articles, and the recommendation diversity can be improved.
The above-described embodiments should not be construed as limiting the scope of the invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions can occur, depending on design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. An item recommendation method, comprising:
generating a first feature vector of the article according to the article information;
generating a first feature vector of a user according to the user information;
determining a user relationship network and an article relationship network;
training a scoring model according to the first feature vector of the article, the first feature vector of the user, the user relationship network and the article relationship network;
and recommending the articles to the user according to the trained scoring model.
2. The method of claim 1,
the generating a first feature vector of the article according to the article information includes:
respectively extracting article characteristics from the multi-modal article information;
and splicing the article features of the article information of each mode to obtain a first feature vector of the article.
3. The method of claim 2,
the modalities include: a text;
the extracting of the article features from the multimodal article information, respectively, comprises:
inputting the article information into word2vec to obtain a word vector;
inputting the word vector into a trained first convolution neural network to obtain the article characteristic;
and/or the presence of a gas in the gas,
the modalities include: a picture;
the extracting of the article features from the multimodal article information, respectively, comprises:
and inputting the article information into a trained second convolutional neural network to obtain the article characteristics.
4. The method of claim 1,
the user information comprises: attribute information and a browsing item sequence;
the generating a first feature vector of the user according to the user information includes:
inputting the attribute information into an embedding layer to obtain a second feature vector of the user;
and inputting the second characteristic vector of the user and the first characteristic vector of each browsed article in the browsed article sequence into a trained vector model to obtain the first characteristic vector of the user.
5. The method of claim 4,
the inputting the second feature vector of the user and the first feature vector of each browsed article in the browsed article sequence into a trained vector model to obtain the first feature vector of the user includes:
respectively inputting the first feature vector of each browsed article into a gate control circulation unit to obtain a context vector of each browsed article;
inputting the context vector of each of the viewed items and the second feature vector of the user into an attention layer, so that the attention layer calculates a weight of each of the viewed items according to the context vector of each of the viewed items and the second feature vector of the user; determining a first feature vector of the user based on the context vector and the weight of each of the viewed items.
6. The method of claim 1,
the determining an item relationship network comprises:
calculating the similarity of the two articles according to the positions of the articles;
generating the article relation network by taking articles as nodes and the similarity of the two articles as edges; wherein, in the item relationship network, if the similarity of two items is larger than a specified threshold, an edge exists between the items.
7. The method of any one of claims 1-6,
the training a scoring model according to the first feature vector of the item, the first feature vector of the user, the user relationship network, and the item relationship network includes:
generating a user adjacency matrix according to the user relationship network;
generating an article adjacency matrix according to the article relation network;
inputting the first eigenvector of the user and the user adjacency matrix into a first graph neural network to obtain a third eigenvector of the user;
inputting the first feature vector of the article and the article adjacency matrix into a second graph neural network to obtain a second feature vector of the article;
inputting the third feature vector of the user and the second feature vector of the article into a multilayer perceptron to obtain a prediction score of the article;
and adjusting parameters of the scoring model according to the prediction score of the article and a preset loss function.
8. An item recommendation device, comprising:
the generating module is configured to generate a first feature vector of the article according to the article information; generating a first feature vector of a user according to the user information;
a determination module configured to determine a user relationship network and an item relationship network;
a prediction module configured to train a scoring model based on the first feature vector of the item, the first feature vector of the user, the user relationship network, and the item relationship network;
and the recommending module is configured to recommend the articles to the user according to the trained scoring model.
9. An electronic device, comprising:
one or more processors;
a storage device for storing one or more programs,
when executed by the one or more processors, cause the one or more processors to implement the method of any one of claims 1-7.
10. A computer-readable medium, on which a computer program is stored, which, when being executed by a processor, carries out the method according to any one of claims 1-7.
CN202110119534.6A 2021-01-28 2021-01-28 Article recommendation method and device Active CN113781150B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110119534.6A CN113781150B (en) 2021-01-28 2021-01-28 Article recommendation method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110119534.6A CN113781150B (en) 2021-01-28 2021-01-28 Article recommendation method and device

Publications (2)

Publication Number Publication Date
CN113781150A true CN113781150A (en) 2021-12-10
CN113781150B CN113781150B (en) 2024-10-22

Family

ID=78835573

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110119534.6A Active CN113781150B (en) 2021-01-28 2021-01-28 Article recommendation method and device

Country Status (1)

Country Link
CN (1) CN113781150B (en)

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108874914A (en) * 2018-05-29 2018-11-23 吉林大学 A kind of information recommendation method based on the long-pending and neural collaborative filtering of picture scroll
KR20190061130A (en) * 2017-11-27 2019-06-05 서울대학교산학협력단 Explainable and accurate recommender method and system using social network information and rating information
WO2019183191A1 (en) * 2018-03-22 2019-09-26 Michael Bronstein Method of news evaluation in social media networks
CN110458627A (en) * 2019-08-19 2019-11-15 华南师范大学 A kind of commodity sequence personalized recommendation method of user oriented preference of dynamic
CN110473042A (en) * 2018-05-11 2019-11-19 北京京东尚科信息技术有限公司 For obtaining the method and device of information
WO2020156389A1 (en) * 2019-01-30 2020-08-06 北京字节跳动网络技术有限公司 Information pushing method and device
CN111523047A (en) * 2020-04-13 2020-08-11 中南大学 Multi-relation collaborative filtering algorithm based on graph neural network
CN111859166A (en) * 2020-07-28 2020-10-30 重庆邮电大学 Article scoring prediction method based on improved graph convolution neural network
CN111881342A (en) * 2020-06-23 2020-11-03 北京工业大学 Recommendation method based on graph twin network
CN111967924A (en) * 2019-05-20 2020-11-20 北京京东尚科信息技术有限公司 Commodity recommendation method, commodity recommendation device, computer device, and medium

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20190061130A (en) * 2017-11-27 2019-06-05 서울대학교산학협력단 Explainable and accurate recommender method and system using social network information and rating information
WO2019183191A1 (en) * 2018-03-22 2019-09-26 Michael Bronstein Method of news evaluation in social media networks
CN110473042A (en) * 2018-05-11 2019-11-19 北京京东尚科信息技术有限公司 For obtaining the method and device of information
CN108874914A (en) * 2018-05-29 2018-11-23 吉林大学 A kind of information recommendation method based on the long-pending and neural collaborative filtering of picture scroll
WO2020156389A1 (en) * 2019-01-30 2020-08-06 北京字节跳动网络技术有限公司 Information pushing method and device
CN111967924A (en) * 2019-05-20 2020-11-20 北京京东尚科信息技术有限公司 Commodity recommendation method, commodity recommendation device, computer device, and medium
CN110458627A (en) * 2019-08-19 2019-11-15 华南师范大学 A kind of commodity sequence personalized recommendation method of user oriented preference of dynamic
CN111523047A (en) * 2020-04-13 2020-08-11 中南大学 Multi-relation collaborative filtering algorithm based on graph neural network
CN111881342A (en) * 2020-06-23 2020-11-03 北京工业大学 Recommendation method based on graph twin network
CN111859166A (en) * 2020-07-28 2020-10-30 重庆邮电大学 Article scoring prediction method based on improved graph convolution neural network

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
IGNACIO GATTI等: "A Hybrid Approach for Artwork Recommendation", 《2019 IEEE SECOND INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE AND KNOWLEDGE ENGINEERING (AIKE)》, 8 August 2019 (2019-08-08), pages 281 - 284 *
SHI, SY (SHI, SHAOYUN)等: "WG4Rec: Modeling Textual Content withWord Graph for News Recommendation", 《PROCEEDINGS OF THE 30TH ACM INTERNATIONAL CONFERENCE ON INFORMATION & KNOWLEDGE MANAGEMENT, CIKM 2021》, 1 January 2021 (2021-01-01), pages 1651 - 1660 *
何昊晨;张丹红;: "基于多维社交关系嵌入的深层图神经网络推荐方法", 《计算机应用》, no. 10, pages 2795 - 2803 *
吴国栋等: "图神经网络推荐研究进展", 《智能系统学报》, no. 1, 29 February 2020 (2020-02-29), pages 14 - 24 *
蔡崇超等: "基于社交网络的推荐系统研究", 《软件导刊》, no. 01, 2 January 2020 (2020-01-02), pages 52 - 55 *
韦鹏程等: "《大数据巨量分析与机器学习的整合与开发》", 31 May 2017, 电子科技大学出版社, pages: 87 *

Also Published As

Publication number Publication date
CN113781150B (en) 2024-10-22

Similar Documents

Publication Publication Date Title
US11514333B2 (en) Combining machine-learning and social data to generate personalized recommendations
CN108604315B (en) Identifying entities using deep learning models
JP6145576B2 (en) Large page recommendation in online social networks
US10395179B2 (en) Methods and systems of venue inference for social messages
US20180101540A1 (en) Diversifying Media Search Results on Online Social Networks
US10699320B2 (en) Marketplace feed ranking on online social networks
CN108629224A (en) Information demonstrating method and device
CN109145280A (en) The method and apparatus of information push
US11263664B2 (en) Computerized system and method for augmenting search terms for increased efficiency and effectiveness in identifying content
US20210406324A1 (en) System and method for providing a content item based on computer vision processing of images
US11430049B2 (en) Communication via simulated user
CN107832338B (en) Method and system for recognizing core product words
TW201503021A (en) Systems and methods for instant e-coupon distribution
US20160012130A1 (en) Aiding composition of themed articles about popular and novel topics and offering users a navigable experience of associated content
US11615444B2 (en) Recommending that an entity in an online system create content describing an item associated with a topic having at least a threshold value of a performance metric and to add a tag describing the item to the content
CN110992127A (en) Article recommendation method and device
CN113495991A (en) Recommendation method and device
CN112783468A (en) Target object sorting method and device
US8972436B2 (en) Translation model and method for matching reviews to objects
CN113744002A (en) Method, device, equipment and computer readable medium for pushing information
CN112100507B (en) Object recommendation method, computing device and computer-readable storage medium
CN113781150B (en) Article recommendation method and device
CN113327145B (en) Article recommendation method and device
CN113762992B (en) Method and device for processing data
EP3306555A1 (en) Diversifying media search results on online social networks

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