CN110059262A - A kind of construction method and device, item recommendation method of the project recommendation model based on hybrid neural networks - Google Patents
A kind of construction method and device, item recommendation method of the project recommendation model based on hybrid neural networks Download PDFInfo
- Publication number
- CN110059262A CN110059262A CN201910318047.5A CN201910318047A CN110059262A CN 110059262 A CN110059262 A CN 110059262A CN 201910318047 A CN201910318047 A CN 201910318047A CN 110059262 A CN110059262 A CN 110059262A
- Authority
- CN
- China
- Prior art keywords
- item
- user
- project
- information
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 61
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000010276 construction Methods 0.000 title claims abstract description 8
- 239000013598 vector Substances 0.000 claims abstract description 68
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 32
- 239000011159 matrix material Substances 0.000 claims description 67
- 230000003993 interaction Effects 0.000 claims description 57
- 230000006870 function Effects 0.000 claims description 39
- 230000004927 fusion Effects 0.000 claims description 22
- 238000012552 review Methods 0.000 claims description 22
- 230000001537 neural effect Effects 0.000 claims description 20
- 238000001914 filtration Methods 0.000 claims description 17
- 238000007781 pre-processing Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 7
- 230000002457 bidirectional effect Effects 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 12
- 230000010354 integration Effects 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 7
- 238000004590 computer program Methods 0.000 description 5
- 238000000354 decomposition reaction Methods 0.000 description 5
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2216/00—Indexing scheme relating to additional aspects of information retrieval not explicitly covered by G06F16/00 and subgroups
- G06F2216/03—Data mining
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The construction method and device, item recommendation method of the invention discloses a kind of project recommendation model based on hybrid neural networks, construction method therein is first filtered comment information, and filtered comment information is pre-processed, then learn contextual feature relevant to project in pretreated comment information, user characteristics and item characteristic in score information using convolutional neural networks；Then the item characteristic in user-project score information is merged and is interacted with contextual feature with comment information, the user characteristics after study are integrated into multi-task learning frame with fused item characteristic again, joint training obtains the project recommendation model based on hybrid neural networks.The present invention is by, into a unified model, can more accurately learn the implicit features vector of user and project for score information and both isomeric data integrations of comment information, to achieve the purpose that improve recommender system performance, improve recommendation effect.
Description
Technical Field
The invention relates to the technical field of data mining, in particular to a method and a device for constructing a project recommendation model based on a hybrid neural network and a project recommendation method.
Background
The development of network information technology meets the information requirement of people and causes the problem of information overload. In the face of massive information, users have difficulty or need to consume large cost to find information which is interesting to the users. Therefore, how to quickly and efficiently select information interested by oneself from massive information has become a big problem in the information age. The recommendation system is produced as a main method for solving the information overload problem, and the recommendation system is used for providing personalized information, products or services for users by analyzing the historical activity information of the users and mining the preferences of the users, so that the personalized requirements of the users are met, and the information overload problem is effectively reduced. In the research of a recommendation system, a collaborative filtering algorithm is the most widely applied recommendation technology at present. The collaborative filtering algorithm predicts the future preference of the user by analyzing the historical feedback information of the user. However, the collaborative filtering algorithm has the problems of serious data sparseness, cold start and the like.
The inventor of the present application finds that the method of the prior art has at least the following technical problems in the process of implementing the present invention:
in recent years, some researchers have utilized review information to reduce problems such as sparseness of data, cold start, and the like. Early recommendation systems utilizing review information mostly focused on mining the subject of the review information using a subject model, and a commonly used subject model was the Dirichlet Allocation model (LDA). For example, relevant research models topics implicit in comment texts by using topic models, combines score information and text information to provide interpretability for quantitative scores recommended by the score information, and then realizes user-item interaction by using expanded matrix decomposition. The text information is introduced through the topic model, so that the recommendation performance is improved to a certain extent. However, this model generally processes comment information using a bag-of-words model, ignoring semantic context information of the comment information. In particular, when the data is too sparse, the latent feature representation learned from the review information by the LDA model may not be very efficient and perform unsatisfactorily. Secondly, fusion of comment information and score information is achieved by expanding a traditional matrix decomposition model for recommendation tasks, and interaction of features is modeled in a linear mode, which is not enough for capturing non-linearity and complex inherent structures of real-world data.
The advantages of deep learning in feature extraction have been achieved with tremendous success in the fields of computer vision, images, natural language processing, and the like. The method for learning the comment information in the hybrid recommendation by using the deep neural network model has become a hot point in the research of a recommendation system. In the related research, a stack denoising automatic encoder is used for learning effective deep feature representation of comment information, a probability graph model is used for integrating text information features and scoring information, and then scoring prediction is realized through a probability matrix. However, when learning the potential features of the comment information using a stacked denoising autoencoder, some noise criteria need to be designed to corrupt the original input. On the basis of word2vec, Kim et al uses a multi-channel idea to set a plurality of groups of different convolution windows, flexibly obtains a plurality of context characteristics of sentences, and applies the texts to sentence classification after expressing the texts into vectors. On the basis, a learner proposes a ConvMF model, learns effective characteristics of the comment information by using a convolutional neural network, and realizes fusion of the comment information and the scoring information through a probability matrix decomposition model. DeepCoNN uses two parallel convolutional neural networks to model user behavior and item attributes from the review information, and at the last level, a decomposition engine is used to capture their interactions for rating prediction.
In the research, the text features are learned through different neural networks, and the comment information is modeled. However, key factors for collaborative filtering recommendations: the feature interaction still adopts a matrix decomposition method instead of the nonlinear interaction of the joint learning features, so that the expression capability of the feature interaction is limited to a certain extent. So that there is a technical problem that the recommendation effect is not good.
Disclosure of Invention
In view of the above, the invention provides a method and an apparatus for constructing a project recommendation model based on a hybrid neural network, and a project recommendation method, so as to solve or at least partially solve the technical problem of poor recommendation effect in the method in the prior art.
The invention provides a method for constructing a project recommendation model based on a hybrid neural network, which comprises the following steps:
step S1: filtering the comment information according to a pre-constructed user-item rating matrix, and preprocessing the filtered comment information, wherein each row in the user-item rating matrix is used for expressing user characteristics in the user-item rating information, and each column in the rating matrix is used for expressing item characteristics in the user-item rating information;
step S2: learning context characteristics related to the items in the preprocessed comment information by using a convolutional neural network, and learning user characteristics and item characteristics in the user-item scoring information by using the convolutional neural network;
step S3: fusing and interacting the project characteristics in the user-project scoring information with the context characteristics related to the project in the comment information to obtain fused project characteristics;
step S4: and integrating the user characteristics learned in the step S2 and the fused project characteristics into a multitask learning framework, and performing combined training to obtain a project recommendation model based on the hybrid neural network.
In one embodiment, the user-item scoring matrix includes item IDs, and step S1 specifically includes:
filtering the comment information containing the item ID in the comment information according to the item ID in the user-item scoring matrix;
and setting the maximum length of the filtered comment information as a preset length, and removing stop words.
In one embodiment, step S2 specifically includes:
step S2.1: for the preprocessed comment information, the comment information is mapped into a word vector matrix based on a word vector model Glove and is used as an initialization parameter input by a neural network, and the initialization parameter is specifically expressed as follows:
D＝[…wi-1wiwi+1…]
wherein, inWhere p denotes each word wiThe embedding dimension of (a) is,l represents item review information CjIs as long as j ∈ [1, M ]]M represents the number of item review messages; then, learning context characteristics related to the project by utilizing convolution and pooling operations of the convolutional neural network;
step S2.2: for user-project scoring information, respectively initializing user features and project features by using a neural network, converting the user features and the project features in the user-project scoring information into dense user feature vectors and project feature vectors, and mapping the user features, the project features and context features related to projects into potential factor vectors through a potential factor matrix in an embedding layer, wherein the potential factor vectors are specifically represented as follows:
wherein,andin order to be a matrix of potential factors,the characteristics of the user are represented by,representing item characteristics, siRepresenting contextual features related to the item.
In one embodiment, the process of learning context features associated with the item using a convolutional neural network specifically includes:
sj＝cnn(W,Cj)
where W represents weights and bias variables to prevent overfitting, CjOriginal item review information, s, representing item jjPotential feature vectors of the comment information of the item j are represented, and an objective function of context features of the item information is learned based on the convolutional neural network and is represented as follows:
wherein v isjInitial value, cnn (W, C) representing a context feature of the itemj) Representing item context features learned through a convolutional neural network, ΘcnnRepresenting the training parameters of the model.
In one embodiment, step S3 specifically includes:
fusing context feature vectors e using element-level multiplicationiAnd item feature vector qiAnd realizing bidirectional interaction of the context characteristics and the project characteristics:
qi＝f(W(qi⊙ei))
wherein,representing the weight matrix, f the activation function of the hidden layer, ⊙ the element-level multiplication.
In one embodiment, the constructed item recommendation model includes a neural interaction layer and a prediction layer, and step S4 specifically includes:
step S4.1: in a neural interaction layer, mapping the user feature vector learned in the user-item scoring information and the fused item feature vector into a final prediction score by using a multilayer perceptron:
wherein,respectively representing the output vector, weight matrix and offset vector of layer l-1, ⊙ representing element-level multiplication, flRepresenting an activation function;
step S4.2: and taking the output of the last layer of the neural interaction layer as the input of a prediction layer, and converting the output into a prediction score as the output of the whole combined model:
wherein,indicates the prediction score, yuiRepresenting a true score;
step S4.3: reduction of prediction scores using mean square error regressionAnd a true score yuiThe objective function of the training process of mean square error regression is:
wherein, wuiA hyperparameter representing the weight of the training instance (u, i),representing model parameters;
step S4.4: and integrating the user characteristics in the step-learned user-scoring information and the fused project characteristics into a multi-task learning framework, and performing combined training to obtain a recommendation model.
In one embodiment, the recommendation model is a linear combination of objective functions of a context feature learning module and a feature interaction module, wherein the feature interaction module comprises a fusion layer, a neural interaction layer and a prediction layer, and the objective function of the joint model is represented as:
a function representing a loss of a predicted rating is expressed,representing a loss function for learning text features using a convolutional neural network,a parameter representing a neural network is determined,wherein theta iscnnParameter, Θ, representing a feature of neural network learning item review informationp,ΘqModel parameters representing a score prediction module.
Based on the same inventive concept, a second aspect of the present invention provides a device for constructing a hybrid neural network-based item recommendation model, including:
the comment information preprocessing module is used for filtering comment information according to a pre-constructed user-item rating matrix and preprocessing the filtered comment information, wherein each row in the user-item rating matrix is used for expressing user characteristics in the user-item rating information, and each column in the rating matrix is used for expressing item characteristics in the user-item rating information;
the feature learning module is used for learning context features related to the items in the preprocessed comment information by using a convolutional neural network and learning user features and item features in the user-item scoring information by using the convolutional neural network;
the feature fusion module is used for fusing and interacting the project features in the user-project scoring information with the context features related to the projects in the comment information to obtain fused project features;
and the project recommendation model construction module is used for integrating the user characteristics learned in the characteristic learning module and the fused project characteristics into a multitask learning frame and performing combined training to obtain a project recommendation model based on the hybrid neural network.
In one embodiment, the user-item scoring matrix includes item IDs, and the data preprocessing and partitioning module is specifically configured to:
filtering the comment information containing the item ID in the comment information according to the item ID in the user-item scoring matrix;
and setting the maximum length of the filtered comment information as a preset length, and removing stop words.
Based on the same inventive concept, a third aspect of the present invention provides an item recommendation method, comprising:
and inputting the comment information and the score information corresponding to the item to be recommended into the item recommendation model based on the hybrid neural network constructed in the first aspect, and obtaining a recommendation result.
One or more technical solutions in the embodiments of the present application have at least one or more of the following technical effects:
the invention provides a method for constructing a project recommendation model based on a hybrid neural network, which comprises the steps of filtering comment information according to a pre-constructed user-project rating matrix, preprocessing the filtered comment information, learning context characteristics related to projects in the preprocessed comment information by using a convolutional neural network, and learning user characteristics and project characteristics in the user-project rating information by using the convolutional neural network; then, fusing and interacting the project characteristics in the user-project scoring information and the context characteristics related to the project in the comment information to obtain fused project characteristics; and integrating the learned user characteristics and the fused project characteristics into a multi-task learning framework, and performing combined training to obtain a project recommendation model based on the hybrid neural network. And a project recommendation method is provided based on the project recommendation model constructed by the construction method.
Compared with the existing method, the method has the advantages that the potential feature vectors of the users and the items are learned from the scoring information and the comment information of the users by utilizing the deep learning related model, and the two heterogeneous data of the scoring information and the comment information are integrated into a unified model, so that the implicit feature vectors of the users and the items can be more accurately learned, and the purposes of improving the performance of a recommendation system and improving the recommendation effect are achieved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flowchart of a method for constructing a hybrid neural network-based project recommendation model according to an embodiment of the present invention;
FIG. 2 is a block diagram of an apparatus for constructing a hybrid neural network-based project recommendation model according to an embodiment of the present invention;
fig. 3 is a schematic flow chart of item recommendation using an item recommendation model based on hybrid neural network fusion review information and scoring information in a specific example.
Detailed Description
The invention aims to provide a method and a device for constructing a project recommendation model based on a hybrid neural network and a project recommendation method, which are used for solving or at least partially solving the technical problem of poor recommendation effect in the prior art, and achieving the technical effects of improving the performance of a recommendation system and improving the recommendation effect.
In order to achieve the technical effects, the main concept of the invention is as follows:
learning the context characteristics of the comment information by using a convolutional neural network, and realizing the interaction of the context characteristics, the item characteristics and the user characteristics and the item characteristics by utilizing the nonlinearity of the neural network; and finally, different data are mapped to the same hidden space through the fusion layer, so that the problems of data sparsity, cold start and the like can be reduced by acquiring the unified representation of the data, and the recommendation performance of the recommendation system is improved.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example one
The embodiment provides a method for constructing a project recommendation model based on a hybrid neural network, please refer to fig. 1, and the method includes:
step S1: and filtering the comment information according to a pre-constructed user-item rating matrix, and preprocessing the filtered comment information, wherein each row in the user-item rating matrix is used for expressing user characteristics in the user-item rating information, and each column in the rating matrix is used for expressing item characteristics in the user-item rating information.
In particular, the user-item scoring matrix is constructed from the user's previous scores for items. In a recommendation system, a model for predicting user preferences based on a user-item scoring matrix is generally referred to as a latent semantic model. Each row in the scoring matrix is referred to as a preference feature of the user (i.e., a user feature), and each column is referred to as an item feature. For example, user a likes books in mathematics, history, etc., while user B likes books in english, chinese, etc. Thus, user a and user B, and the items: mathematics, history, English and Chinese form a matrix with 2 rows and 4 columns. Each row in the matrix represents the preference of a user for different items, and the preference characteristics of the user are reflected. And each column represents the preference degree of different users for the same item, and embodies the characteristics of the item. Thus, the preference characteristics of the user and the attribute characteristics of the items can be learned from the user-item scoring matrix.
In one embodiment, the user-item scoring matrix includes item IDs, and step S1 specifically includes:
filtering the comment information containing the item ID in the comment information according to the item ID in the user-item scoring matrix;
and setting the maximum length of the filtered comment information as a preset length, and removing stop words.
Specifically, the preset length may be set according to actual conditions, for example, set to 200 words, 250 words, 300 words, and so on.
Step S2: and learning context characteristics related to the items in the preprocessed comment information by using a convolutional neural network, and learning user characteristics and item characteristics in the user-item scoring information by using the convolutional neural network.
Specifically, the comment information can be converted into a word vector matrix by using a Glove word vector model, and then context characteristics of the comment information, which are related to the item, can be learned by using a rolling machine neural network; and for the user-item scoring information, converting the user-item scoring matrix into a low-dimensional dense user feature vector and a low-dimensional dense item feature vector by using a neural network.
In a specific implementation, comment information is referred to. Converting each word in the item comment information into a word vector with a reserved word sequence by using a GloVe (word embedding model) popular in the field of natural language processing to form a word vector matrix of the item comment information, and then learning an internal structure of the comment information by using a convolutional neural network capable of effectively using the word sequence to overcome the influence of word sequence conversion on the context of the auxiliary information; and capturing the deep-level features related to the items by using the modes of shared weight, sub-sampling and the like of the convolutional neural network.
Scoring information for the user-item. The neural network is used for respectively initializing the user characteristics and the item characteristics, learning the optimal initialization parameters of the user and the item in the scoring information, and predicting the preference of the user to the item by using the optimal parameters, so that the performance of the recommendation system is improved.
In one embodiment, step S2 specifically includes:
step S2.1: for the preprocessed comment information, the comment information is mapped into a word vector matrix based on a word vector model Glove and is used as an initialization parameter input by a neural network, and the initialization parameter is specifically expressed as follows:
D＝[…wi-1wiwi+1…]
wherein, inWhere p denotes each word wiThe embedding dimension of (a) is,l represents item review information CjIs as long as j ∈ [1, M ]]M represents the number of item review messages; then, learning context characteristics related to the project by utilizing convolution and pooling operations of the convolutional neural network;
step S2.2: for user-project scoring information, respectively initializing user features and project features by using a neural network, converting the user features and the project features in the user-project scoring information into dense user feature vectors and project feature vectors, and mapping the user features, the project features and context features related to projects into potential factor vectors through a potential factor matrix in an embedding layer, wherein the potential factor vectors are specifically represented as follows:
wherein,andin order to be a matrix of potential factors,the characteristics of the user are represented by,representing item characteristics, siRepresenting contextual features related to the item.
Specifically, when learning context features related to an item, the learned context features are mapped to a dimension as a final input at the output layer.
The process of learning the context features related to the items by using the convolutional neural network specifically includes:
sj＝cnn(W,Cj)
where W represents weights and bias variables to prevent overfitting, CjOriginal item review information, s, representing item jjPotential feature vectors of the comment information of the item j are represented, and an objective function of context features of the item information is learned based on the convolutional neural network and is represented as follows:
wherein v isjInitial value, cnn (W, C) representing a context feature of the itemj) Representing item context features learned through a convolutional neural network, ΘcnnRepresenting the training parameters of the model.
The invention optimizes the network by using random gradient descent, thereby obtaining the contextual characteristics of the item comment information.
Step S3: and fusing and interacting the project characteristics in the user-project scoring information and the context characteristics related to the project in the comment information to obtain fused project characteristics.
In particular, fusion and interaction of project features in user-project scoring information and context features related to projects in comment information can be achieved by adding a fusion layer.
In one embodiment, step S3 specifically includes:
fusing context feature vectors e using element-level multiplicationiAnd item feature vector qiAnd realizing bidirectional interaction of the context characteristics and the project characteristics:
qi＝f(W(qi⊙ei))
wherein,representing the weight matrix, f the activation function of the hidden layer, ⊙ the element-level multiplication.
Specifically, the present embodiment realizes fusion of heterogeneous information features by fusing the context features in the comment information on the item learned by the convolutional neural network with the item features obtained in step S3.
Step S4: and integrating the user characteristics learned in the step S2 and the fused project characteristics into a multitask learning framework, and performing combined training to obtain a project recommendation model based on the hybrid neural network.
Specifically, the user feature vector learned in step S2 and the item features fused in step S3 are used as initialization parameters of final user features and item features, multiple interactions of the user features and the item features are realized by using a multilayer perceptual neural network, and a final predicted value of the user-item score is obtained through training. And taking the constructed user characteristics and the fused project characteristics as final user characteristics and project characteristic initialization parameters, realizing multiple interactions of the user characteristics and the project characteristics by utilizing a multilayer perception neural network, and training to obtain a final user-project score prediction value.
In one embodiment, the constructed item recommendation model includes a neural interaction layer and a prediction layer, and step S4 specifically includes:
step S4.1: in a neural interaction layer, mapping the user feature vector learned in the user-item scoring information and the fused item feature vector into a final prediction score by using a multilayer perceptron:
wherein,respectively representing the output vector, weight matrix and offset vector of layer l-1, ⊙ representing element-level multiplication, flRepresenting an activation function;
step S4.2: and taking the output of the last layer of the neural interaction layer as the input of a prediction layer, and converting the output into a prediction score as the output of the whole combined model:
wherein,indicates the prediction score, yuiRepresenting a true score;
step S4.3: reduction of prediction scores using mean square error regressionAnd a true score yuiThe objective function of the training process of mean square error regression is:
wherein, wuiA hyperparameter representing the weight of the training instance (u, i),representing model parameters;
step S4.4: and integrating the user characteristics in the learned user-scoring information and the fused project characteristics into a multi-task learning framework, and performing combined training to obtain a recommendation model.
Specifically, in step S4.1, nerves are used in the nerve interaction layerThe non-linearity of the network enables close coupling of textual information and scoring information and non-linear interaction of features. f. oflExpressing an activation function, a Rectifier unit (ReLU) that is more biological and less prone to saturation is used as a nonlinear activation function, and ReLU (x) max (x, 0). In order to realize final scoring prediction, learning of context features of scoring information and item description information is integrated into a unified multi-task learning framework for joint training, and a recommendation model based on mixed neural network fusion text information and scoring information is obtained.
The recommendation model is a linear combination of objective functions of a context feature learning module and a feature interaction module, wherein the feature interaction module comprises a fusion layer, a neural interaction layer and a prediction layer, and the objective function of the combined model is expressed as:
a function representing a loss of a predicted rating is expressed,representing a loss function for learning text features using a convolutional neural network,a parameter representing a neural network is determined,wherein theta iscnnParameter, Θ, representing a feature of neural network learning item review informationp,ΘqModel parameters representing a score prediction module.
In particular, forUsing classical convolution godsLearning context characteristics of the item comment information through a network model; for theTo learn a complex structure of user-item interactions using a neural-aware network to enhance the non-linearity of the user-item interactions. The optimization of the objective function is realized by jointly training the two loss functions, and the prediction of the user preference is realized. The final prediction model can be expressed as:
a matrix of latent factors representing users and items, respectively, each row representing a user and an item, respectively, with the purpose of converting user and item features into a latent feature vector, Θp,ΘqRepresenting parameters of the embedding layer and parameters of the prediction layer, respectively. f represents the interaction function of the model. To increase the non-linearity of the user-item interaction, the depth of the model is increased by stacking multiple layers of perceptrons, and thus can be expressed as:
wherein phi1,φ2,…,φn-1,φn1 to n layers of a multilayer perceptron representing a neural network, of which phinRepresenting the output layer, the number of layers of the sensor is n.
Generally, compared with the prior art, the technical scheme adopted by the invention has the following technical effects:
1. a potential feature representation of review information related to the item is automatically obtained using a convolutional neural network. Deep-level features related to the items can be effectively captured through ways of sharing weight, sub-sampling and the like, and the influence of word order and context information on the extracted potential features of the items can be considered at the same time, so that potential feature representation better than that of a shallow model is generated;
2. deep learning performs automatic feature learning from multi-source heterogeneous data, so that different data are mapped to the same hidden space, and uniform representation of the data can be obtained. On the basis, the traditional recommendation method is fused for recommendation, multi-source heterogeneous data can be effectively utilized, and the problems of data sparsity and cold start in a traditional recommendation system are solved.
3. In the feature interaction module, the interaction of the context features and the project features is realized through a fusion layer, and then the complexity of the interaction of the user features and the project features can be better learned by utilizing a neural network.
By combining the three points, the method for constructing the project recommendation model based on the hybrid neural network can learn the implicit characteristic vector of the user more accurately by the constructed recommendation module, so that the performance of the recommendation system is improved.
Based on the same inventive concept, the application also provides a device corresponding to the method for constructing the project recommendation model based on the hybrid neural network in the first embodiment, which is detailed in the second embodiment.
Example two
The embodiment provides a device for constructing a project recommendation model based on a hybrid neural network, please refer to fig. 2, the device includes:
the comment information preprocessing module 201 is configured to filter comment information according to a pre-constructed user-item rating matrix, and preprocess the filtered comment information, where each row in the user-item rating matrix is used to represent a user feature in the user-item rating information, and each column in the rating matrix is used to represent an item feature in the user-item rating information;
the feature learning module 202 is configured to learn context features related to the items in the preprocessed comment information by using a convolutional neural network, and learn user features and item features in the user-item scoring information by using the convolutional neural network;
the feature fusion module 203 is configured to fuse and interact a project feature in the user-project scoring information with a context feature related to the project in the comment information, so as to obtain a fused project feature;
and the project recommendation model building module 204 is used for integrating the user characteristics learned in the characteristic learning module and the fused project characteristics into a multitask learning framework, and performing combined training to obtain a project recommendation model based on the hybrid neural network.
In one embodiment, the user-item scoring matrix includes item IDs, and the data preprocessing and partitioning module 201 is specifically configured to:
filtering the comment information containing the item ID in the comment information according to the item ID in the user-item scoring matrix;
and setting the maximum length of the filtered comment information as a preset length, and removing stop words.
In one embodiment, the feature learning module 202 is specifically configured to perform the following steps:
step S2.1: for the preprocessed comment information, the comment information is mapped into a word vector matrix based on a word vector model Glove and is used as an initialization parameter input by a neural network, and the initialization parameter is specifically expressed as follows:
D＝[…wi-1wiwi+1…]
wherein, inWhere p denotes each word wiThe embedding dimension of (a) is,l represents item review information CjIs as long as j ∈ [1, M ]]M represents the number of item review messages; then, learning context characteristics related to the project by utilizing convolution and pooling operations of the convolutional neural network;
step S2.2: for user-project scoring information, respectively initializing user features and project features by using a neural network, converting the user features and the project features in the user-project scoring information into dense user feature vectors and project feature vectors, and mapping the user features, the project features and context features related to projects into potential factor vectors through a potential factor matrix in an embedding layer, wherein the potential factor vectors are specifically represented as follows:
wherein,andin order to be a matrix of potential factors,the characteristics of the user are represented by,representing item characteristics, siRepresenting contextual features related to the item.
In one embodiment, the process of learning context features associated with the item using a convolutional neural network specifically includes:
sj＝cnn(W,Cj)
where W represents weights and bias variables to prevent overfitting, CjOriginal item representing item jComment information, sjPotential feature vectors of the comment information of the item j are represented, and an objective function of context features of the item information is learned based on the convolutional neural network and is represented as follows:
wherein v isjInitial value, cnn (W, C) representing a context feature of the itemj) Representing item context features learned through a convolutional neural network, ΘcnnRepresenting the training parameters of the model.
In one embodiment, the feature fusion module 203 is specifically configured to:
fusing context feature vectors e using element-level multiplicationiAnd item feature vector qiAnd realizing bidirectional interaction of the context characteristics and the project characteristics:
qi＝f(W(qi⊙ei))
wherein,representing the weight matrix, f the activation function of the hidden layer, ⊙ the element-level multiplication.
In one embodiment, the constructed item recommendation model includes a neural interaction layer and a prediction layer, and the item recommendation model construction module 204 is specifically configured to perform the following steps:
step S4.1: in a neural interaction layer, mapping the user feature vector learned in the user-item scoring information and the fused item feature vector into a final prediction score by using a multilayer perceptron:
wherein,respectively representing the output vector, weight matrix and offset vector of layer l-1, ⊙ representing element-level multiplication, flRepresenting an activation function;
step S4.2: and taking the output of the last layer of the neural interaction layer as the input of a prediction layer, and converting the output into a prediction score as the output of the whole combined model:
wherein,indicates the prediction score, yuiRepresenting a true score;
step S4.3: reduction of prediction scores using mean square error regressionAnd a true score yuiThe objective function of the training process of mean square error regression is:
wherein, wuiA hyperparameter representing the weight of the training instance (u, i),representing model parameters;
step S4.4: and integrating the user characteristics in the step-learned user-scoring information and the fused project characteristics into a multi-task learning framework, and performing combined training to obtain a recommendation model.
In one embodiment, the recommendation model is a linear combination of objective functions of a context feature learning module and a feature interaction module, wherein the feature interaction module comprises a fusion layer, a neural interaction layer and a prediction layer, and the objective function of the joint model is represented as:
a function representing a loss of a predicted rating is expressed,representing a loss function for learning text features using a convolutional neural network,a parameter representing a neural network is determined,wherein theta iscnnParameter, Θ, representing a feature of neural network learning item review informationp,ΘqModel parameters representing a score prediction module.
Since the apparatus described in the second embodiment of the present invention is an apparatus used for implementing the method for constructing the hybrid neural network-based project recommendation model in the first embodiment of the present invention, a person skilled in the art can understand the specific structure and the deformation of the apparatus based on the method described in the first embodiment of the present invention, and thus, the details are not described herein. All the devices adopted in the method of the first embodiment of the present invention belong to the protection scope of the present invention.
Based on the same inventive concept, the application also provides a project recommendation method based on the hybrid neural network project recommendation model established in the first embodiment, and the detailed description is shown in the third embodiment.
EXAMPLE III
The embodiment provides an item recommendation method, which comprises the following steps:
and inputting the comment information and the score information corresponding to the item to be recommended into a constructed item recommendation model based on the hybrid neural network to obtain a recommendation result.
To more clearly illustrate the item recommendation method of the present invention, a specific process is described below, please refer to fig. 1. The project recommendation model constructed by the invention comprises an embedding layer, a convolution layer, a pooling layer, a fusion layer, a connection layer, a neural interaction layer and a prediction layer.
Firstly, collecting training data, judging whether the training data is unstructured text data, if so, obtaining a preprocessed project description document set, and if not, further judging whether the training data is structured project features.
After the preprocessed project description document set is obtained, a project description document is randomly extracted from the project description document set and converted into a word vector matrix, the context features of the project description document are extracted through a convolutional layer of a convolutional neural network, then the context features of the project description document can be most represented through the pooling layer of the convolutional neural network, and then the pooling sampling is continued.
When the structured item features are judged, the item features are converted into dense item feature vectors through the embedding layer, and when the structured item features are not judged, the item features are converted into dense user feature vectors through the embedding layer.
And the context characteristics of the extracted project description document are fused with the project characteristics through a fusion layer. And fusion of the context characteristics and the project characteristics of the project comment information is realized through the fusion layer, and interaction of the context characteristics and the project characteristics is promoted.
And finally, the output of the last layer of the neural interaction layer is used as the input of the prediction layer and is converted into a prediction score to be used as the output of the whole combined model through the functions of the connection layer and the neural interaction layer, so that the project recommendation is realized.
Generally speaking, the invention discloses a project recommendation method realized by a project recommendation model based on a hybrid neural network, which mainly solves two problems of negligence of a collaborative filtering recommendation algorithm based on comment information: (1) how to obtain a more effective comment information representation; (2) how to capture the complexity of feature interactions during the feature interaction phase. The invention defines a hybrid neural network framework, which is mainly composed of two components: a contextual feature learning component and a feature interaction component. And the context feature learning component is used for capturing the context features of the comment information by utilizing a convolutional neural network. And the feature interaction component fuses various features through a fusion layer, then updates the parameters of the two components through a nonlinear learning feature interaction complex structure of a multilayer perception neural network, and finally provides personalized recommendation for the user by using a joint training method to achieve the prediction score with the minimum error. The method can effectively capture the context characteristics of the comment information to reduce the sparsity of data, and utilizes the nonlinearity of the neural network to capture the complexity of characteristic interaction, thereby improving the recommendation effect.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various modifications and variations can be made in the embodiments of the present invention without departing from the spirit or scope of the embodiments of the invention. Thus, if such modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to encompass such modifications and variations.
Claims (10)
1. A method for constructing a project recommendation model based on a hybrid neural network is characterized by comprising the following steps:
step S1: filtering the comment information according to a pre-constructed user-item rating matrix, and preprocessing the filtered comment information, wherein each row in the user-item rating matrix is used for expressing user characteristics in the user-item rating information, and each column in the rating matrix is used for expressing item characteristics in the user-item rating information;
step S2: learning context characteristics related to the items in the preprocessed comment information by using a convolutional neural network, and learning user characteristics and item characteristics in the user-item scoring information by using the convolutional neural network;
step S3: fusing and interacting the project characteristics in the user-project scoring information with the context characteristics related to the project in the comment information to obtain fused project characteristics;
step S4: and integrating the user characteristics learned in the step S2 and the fused project characteristics into a multitask learning framework, and performing combined training to obtain a project recommendation model based on the hybrid neural network.
2. The method of claim 1, wherein the user-item scoring matrix includes item IDs, and step S1 specifically includes:
filtering the comment information containing the item ID in the comment information according to the item ID in the user-item scoring matrix;
and setting the maximum length of the filtered comment information as a preset length, and removing stop words.
3. The method according to claim 1, wherein step S2 specifically comprises:
step S2.1: for the preprocessed comment information, the comment information is mapped into a word vector matrix based on a word vector model Glove and is used as an initialization parameter input by a neural network, and the initialization parameter is specifically expressed as follows:
D＝[… wi-1wiwi+1…]
wherein, inWhere p denotes each word wiThe embedding dimension of (a) is,l represents item review information CjIs as long as j ∈ [1, M ]]M represents the number of item review messages; and then using convolutional neural networksLearning context features associated with the items by convolution and pooling operations;
step S2.2: for user-project scoring information, respectively initializing user features and project features by using a neural network, converting the user features and the project features in the user-project scoring information into dense user feature vectors and project feature vectors, and mapping the user features, the project features and context features related to projects into potential factor vectors through a potential factor matrix in an embedding layer, wherein the potential factor vectors are specifically represented as follows:
wherein,andin order to be a matrix of potential factors,the characteristics of the user are represented by,representing item characteristics, siRepresenting contextual features related to the item.
4. The method of claim 3, wherein learning the context features associated with the items using a convolutional neural network specifically comprises:
sj＝cnn(W,Cj)
where W represents weights and bias variables to prevent overfitting, CjOriginal item review information, s, representing item jjPotential feature vectors of the comment information of the item j are represented, and an objective function of context features of the item information is learned based on the convolutional neural network and is represented as follows:
wherein v isjInitial value, cnn (W, C) representing a context feature of the itemj) Representing item context features learned through a convolutional neural network, ΘcnnRepresenting the training parameters of the model.
5. The method according to claim 1, wherein step S3 specifically comprises:
fusing context feature vectors e using element-level multiplicationiAnd item feature vector qiAnd realizing bidirectional interaction of the context characteristics and the project characteristics:
qi＝f(W(qi⊙ei))
wherein,representing the weight matrix, f the activation function of the hidden layer, ⊙ the element-level multiplication.
6. The method of claim 1, wherein the constructed item recommendation model includes a neural interaction layer and a prediction layer, and step S4 specifically includes:
step S4.1: in a neural interaction layer, mapping the user feature vector learned in the user-item scoring information and the fused item feature vector into a final prediction score by using a multilayer perceptron:
wherein,respectively representing the output vector, weight matrix and offset vector of layer l-1, ⊙ representing element-level multiplication, flExpress laserA live function;
step S4.2: and taking the output of the last layer of the neural interaction layer as the input of a prediction layer, and converting the output into a prediction score as the output of the whole combined model:
wherein,indicates the prediction score, yuiRepresenting a true score;
step S4.3: reduction of prediction scores using mean square error regressionAnd a true score yuiThe objective function of the training process of mean square error regression is:
wherein, wuiA hyperparameter representing the weight of the training instance (u, i),representing model parameters;
step S4.4: and integrating the user characteristics in the step-learned user-scoring information and the fused project characteristics into a multi-task learning framework, and performing combined training to obtain a recommendation model.
7. The method of claim 6, wherein the recommendation model is a linear combination of objective functions of a context feature learning module and a feature interaction module, wherein the feature interaction module comprises a fusion layer, a neural interaction layer, and a prediction layer, and the objective function of the joint model is represented as:
a function representing a loss of a predicted rating is expressed,representing a loss function for learning text features using a convolutional neural network,a parameter representing a neural network is determined,wherein theta iscnnParameter, Θ, representing a feature of neural network learning item review informationp,ΘqModel parameters representing a score prediction module.
8. A device for constructing a hybrid neural network-based project recommendation model is characterized by comprising the following components:
the comment information preprocessing module is used for filtering comment information according to a pre-constructed user-item rating matrix and preprocessing the filtered comment information, wherein each row in the user-item rating matrix is used for expressing user characteristics in the user-item rating information, and each column in the rating matrix is used for expressing item characteristics in the user-item rating information;
the feature learning module is used for learning context features related to the items in the preprocessed comment information by using a convolutional neural network and learning user features and item features in the user-item scoring information by using the convolutional neural network;
the feature fusion module is used for fusing and interacting the project features in the user-project scoring information with the context features related to the projects in the comment information to obtain fused project features;
and the project recommendation model construction module is used for integrating the user characteristics learned in the characteristic learning module and the fused project characteristics into a multitask learning frame and performing combined training to obtain a project recommendation model based on the hybrid neural network.
9. The apparatus of claim 8, wherein the user-item scoring matrix comprises item IDs, and the data preprocessing and partitioning module is specifically configured to:
filtering the comment information containing the item ID in the comment information according to the item ID in the user-item scoring matrix;
and setting the maximum length of the filtered comment information as a preset length, and removing stop words.
10. An item recommendation method, comprising:
inputting the comment information and the score information corresponding to the item to be recommended into the item recommendation model based on the hybrid neural network constructed according to any one of claims 1 to 7, and obtaining a recommendation result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910318047.5A CN110059262B (en) | 2019-04-19 | 2019-04-19 | Project recommendation model construction method and device based on hybrid neural network and project recommendation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910318047.5A CN110059262B (en) | 2019-04-19 | 2019-04-19 | Project recommendation model construction method and device based on hybrid neural network and project recommendation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110059262A true CN110059262A (en) | 2019-07-26 |
| CN110059262B CN110059262B (en) | 2021-07-02 |
Family
ID=67319697
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910318047.5A Expired - Fee Related CN110059262B (en) | 2019-04-19 | 2019-04-19 | Project recommendation model construction method and device based on hybrid neural network and project recommendation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110059262B (en) |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110517069A (en) * | 2019-08-15 | 2019-11-29 | 中森云链(成都)科技有限责任公司 | A kind of multi-modal user's portrait method and system |
| CN111079015A (en) * | 2019-12-17 | 2020-04-28 | 腾讯科技（深圳）有限公司 | Recommendation method and device, computer equipment and storage medium |
| CN111104789A (en) * | 2019-11-22 | 2020-05-05 | 华中师范大学 | Text scoring method, device and system |
| CN111179031A (en) * | 2019-12-23 | 2020-05-19 | 第四范式（北京）技术有限公司 | Training method, device and system for commodity recommendation model |
| CN111209475A (en) * | 2019-12-27 | 2020-05-29 | 武汉大学 | Interest point recommendation method and device based on space-time sequence and social embedded ranking |
| CN111274789A (en) * | 2020-02-06 | 2020-06-12 | 支付宝(杭州)信息技术有限公司 | Training method and device of text prediction model |
| CN111368197A (en) * | 2020-03-04 | 2020-07-03 | 哈尔滨理工大学 | Deep learning-based comment recommendation system and method |
| CN111667344A (en) * | 2020-06-08 | 2020-09-15 | 中森云链(成都)科技有限责任公司 | Personalized recommendation method integrating comments and scores |
| CN111782964A (en) * | 2020-06-23 | 2020-10-16 | 北京智能工场科技有限公司 | Community post recommendation method |
| CN111931045A (en) * | 2020-07-30 | 2020-11-13 | 北京邮电大学 | Heterogeneous information network cold start recommendation method and device based on meta-learning |
| CN111967578A (en) * | 2020-08-04 | 2020-11-20 | 厦门大学 | Construction method of depth recommendation system framework based on uncompensated decision mechanism |
| CN112084415A (en) * | 2020-09-17 | 2020-12-15 | 辽宁工程技术大学 | Recommendation method based on analysis of long-term and short-term coupling relationship between user and project |
| CN112100485A (en) * | 2020-08-20 | 2020-12-18 | 齐鲁工业大学 | Comment-based scoring prediction item recommendation method and system |
| CN112328900A (en) * | 2020-11-27 | 2021-02-05 | 北京工业大学 | Deep learning recommendation method integrating scoring matrix and comment text |
| CN112541639A (en) * | 2020-12-22 | 2021-03-23 | 宜宾电子科技大学研究院 | Recommendation system scoring prediction method based on graph neural network and attention mechanism |
| CN112612948A (en) * | 2020-12-14 | 2021-04-06 | 浙大城市学院 | Deep reinforcement learning-based recommendation system construction method |
| CN112860880A (en) * | 2021-03-23 | 2021-05-28 | 辽宁工程技术大学 | Comment text and feature weighting fused four-dimensional tensor decomposition recommendation method |
| CN112966148A (en) * | 2021-03-05 | 2021-06-15 | 安徽师范大学 | Video recommendation method and system based on deep learning and feature fusion |
| CN113312644A (en) * | 2021-06-15 | 2021-08-27 | 杭州金智塔科技有限公司 | Cross-domain recommendation model training method and system based on privacy protection |
| CN113590964A (en) * | 2021-08-04 | 2021-11-02 | 燕山大学 | Deep neural network Top-N recommendation algorithm based on heterogeneous modeling |
| CN114048826A (en) * | 2021-11-30 | 2022-02-15 | 中国建设银行股份有限公司 | Recommendation model training method, device, equipment and medium |
| CN114365465A (en) * | 2019-08-15 | 2022-04-15 | 微软技术许可有限责任公司 | Message recommendation system |
| CN114385926A (en) * | 2021-12-02 | 2022-04-22 | 山东师范大学 | Robustness recommendation method and system for reducing proportion of useless information in attack environment |
| CN115019933A (en) * | 2022-06-16 | 2022-09-06 | 浙江工业大学 | Amblyopia training scheme recommendation method fusing GMF and CDAE |
| US11886971B2 (en) | 2019-08-15 | 2024-01-30 | Microsoft Technology Licensing, Llc | Multiple-entity-based recommendation system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944035A (en) * | 2017-12-13 | 2018-04-20 | 合肥工业大学 | A kind of image recommendation method for merging visual signature and user's scoring |
| CN108090229A (en) * | 2018-01-10 | 2018-05-29 | 广东工业大学 | A kind of method and apparatus that rating matrix is determined based on convolutional neural networks |
| CN108287904A (en) * | 2018-05-09 | 2018-07-17 | 重庆邮电大学 | A kind of document context perception recommendation method decomposed based on socialization convolution matrix |
| US20180218429A1 (en) * | 2017-01-31 | 2018-08-02 | Wal-Mart Stores, Inc. | Systems and methods for utilizing a convolutional neural network architecture for visual product recommendations |
| CN109299396A (en) * | 2018-11-28 | 2019-02-01 | 东北师范大学 | Merge the convolutional neural networks collaborative filtering recommending method and system of attention model |
-
2019
- 2019-04-19 CN CN201910318047.5A patent/CN110059262B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180218429A1 (en) * | 2017-01-31 | 2018-08-02 | Wal-Mart Stores, Inc. | Systems and methods for utilizing a convolutional neural network architecture for visual product recommendations |
| CN107944035A (en) * | 2017-12-13 | 2018-04-20 | 合肥工业大学 | A kind of image recommendation method for merging visual signature and user's scoring |
| CN108090229A (en) * | 2018-01-10 | 2018-05-29 | 广东工业大学 | A kind of method and apparatus that rating matrix is determined based on convolutional neural networks |
| CN108287904A (en) * | 2018-05-09 | 2018-07-17 | 重庆邮电大学 | A kind of document context perception recommendation method decomposed based on socialization convolution matrix |
| CN109299396A (en) * | 2018-11-28 | 2019-02-01 | 东北师范大学 | Merge the convolutional neural networks collaborative filtering recommending method and system of attention model |
Cited By (42)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114365465B (en) * | 2019-08-15 | 2023-09-01 | 微软技术许可有限责任公司 | Systems, methods, and computer-readable media for automatically generating recommendations |
| CN114365465A (en) * | 2019-08-15 | 2022-04-15 | 微软技术许可有限责任公司 | Message recommendation system |
| CN110517069A (en) * | 2019-08-15 | 2019-11-29 | 中森云链(成都)科技有限责任公司 | A kind of multi-modal user's portrait method and system |
| US11886971B2 (en) | 2019-08-15 | 2024-01-30 | Microsoft Technology Licensing, Llc | Multiple-entity-based recommendation system |
| CN111104789A (en) * | 2019-11-22 | 2020-05-05 | 华中师范大学 | Text scoring method, device and system |
| CN111104789B (en) * | 2019-11-22 | 2023-12-29 | 华中师范大学 | Text scoring method, device and system |
| CN111079015A (en) * | 2019-12-17 | 2020-04-28 | 腾讯科技（深圳）有限公司 | Recommendation method and device, computer equipment and storage medium |
| CN111079015B (en) * | 2019-12-17 | 2021-08-31 | 腾讯科技（深圳）有限公司 | Recommendation method and device, computer equipment and storage medium |
| CN111179031A (en) * | 2019-12-23 | 2020-05-19 | 第四范式（北京）技术有限公司 | Training method, device and system for commodity recommendation model |
| CN111179031B (en) * | 2019-12-23 | 2023-09-26 | 第四范式（北京）技术有限公司 | Training method, device and system for commodity recommendation model |
| CN111209475A (en) * | 2019-12-27 | 2020-05-29 | 武汉大学 | Interest point recommendation method and device based on space-time sequence and social embedded ranking |
| CN111209475B (en) * | 2019-12-27 | 2022-03-15 | 武汉大学 | Interest point recommendation method and device based on space-time sequence and social embedded ranking |
| CN111274789A (en) * | 2020-02-06 | 2020-06-12 | 支付宝(杭州)信息技术有限公司 | Training method and device of text prediction model |
| CN111368197B (en) * | 2020-03-04 | 2022-05-27 | 哈尔滨理工大学 | Deep learning-based comment recommendation system and method |
| CN111368197A (en) * | 2020-03-04 | 2020-07-03 | 哈尔滨理工大学 | Deep learning-based comment recommendation system and method |
| CN111667344A (en) * | 2020-06-08 | 2020-09-15 | 中森云链(成都)科技有限责任公司 | Personalized recommendation method integrating comments and scores |
| CN111782964B (en) * | 2020-06-23 | 2024-02-09 | 北京智能工场科技有限公司 | Recommendation method of community posts |
| CN111782964A (en) * | 2020-06-23 | 2020-10-16 | 北京智能工场科技有限公司 | Community post recommendation method |
| CN111931045B (en) * | 2020-07-30 | 2022-11-25 | 北京邮电大学 | Heterogeneous information network cold start recommendation method and device based on meta-learning |
| CN111931045A (en) * | 2020-07-30 | 2020-11-13 | 北京邮电大学 | Heterogeneous information network cold start recommendation method and device based on meta-learning |
| CN111967578B (en) * | 2020-08-04 | 2022-06-21 | 厦门大学 | Construction method of depth recommendation system framework based on uncompensated decision mechanism |
| CN111967578A (en) * | 2020-08-04 | 2020-11-20 | 厦门大学 | Construction method of depth recommendation system framework based on uncompensated decision mechanism |
| CN112100485B (en) * | 2020-08-20 | 2024-07-05 | 齐鲁工业大学 | Comment-based scoring prediction article recommendation method and system |
| CN112100485A (en) * | 2020-08-20 | 2020-12-18 | 齐鲁工业大学 | Comment-based scoring prediction item recommendation method and system |
| CN112084415A (en) * | 2020-09-17 | 2020-12-15 | 辽宁工程技术大学 | Recommendation method based on analysis of long-term and short-term coupling relationship between user and project |
| CN112084415B (en) * | 2020-09-17 | 2024-02-02 | 辽宁工程技术大学 | Recommendation method based on analysis of long-term and short-term time coupling relation between user and project |
| CN112328900A (en) * | 2020-11-27 | 2021-02-05 | 北京工业大学 | Deep learning recommendation method integrating scoring matrix and comment text |
| CN112612948A (en) * | 2020-12-14 | 2021-04-06 | 浙大城市学院 | Deep reinforcement learning-based recommendation system construction method |
| CN112541639B (en) * | 2020-12-22 | 2023-07-07 | 宜宾电子科技大学研究院 | Recommendation system scoring prediction method based on graph neural network and attention mechanism |
| CN112541639A (en) * | 2020-12-22 | 2021-03-23 | 宜宾电子科技大学研究院 | Recommendation system scoring prediction method based on graph neural network and attention mechanism |
| CN112966148A (en) * | 2021-03-05 | 2021-06-15 | 安徽师范大学 | Video recommendation method and system based on deep learning and feature fusion |
| CN112860880A (en) * | 2021-03-23 | 2021-05-28 | 辽宁工程技术大学 | Comment text and feature weighting fused four-dimensional tensor decomposition recommendation method |
| CN112860880B (en) * | 2021-03-23 | 2024-03-08 | 辽宁工程技术大学 | Four-dimensional tensor decomposition recommendation method integrating comment text and feature weighting |
| CN113312644B (en) * | 2021-06-15 | 2022-05-24 | 杭州金智塔科技有限公司 | Cross-domain recommendation model training method and system based on privacy protection |
| CN113312644A (en) * | 2021-06-15 | 2021-08-27 | 杭州金智塔科技有限公司 | Cross-domain recommendation model training method and system based on privacy protection |
| CN113590964A (en) * | 2021-08-04 | 2021-11-02 | 燕山大学 | Deep neural network Top-N recommendation algorithm based on heterogeneous modeling |
| CN113590964B (en) * | 2021-08-04 | 2023-05-23 | 燕山大学 | Deep neural network Top-N recommendation method based on heterogeneous modeling |
| CN114048826B (en) * | 2021-11-30 | 2024-04-30 | 中国建设银行股份有限公司 | Recommendation model training method, device, equipment and medium |
| CN114048826A (en) * | 2021-11-30 | 2022-02-15 | 中国建设银行股份有限公司 | Recommendation model training method, device, equipment and medium |
| CN114385926A (en) * | 2021-12-02 | 2022-04-22 | 山东师范大学 | Robustness recommendation method and system for reducing proportion of useless information in attack environment |
| CN115019933A (en) * | 2022-06-16 | 2022-09-06 | 浙江工业大学 | Amblyopia training scheme recommendation method fusing GMF and CDAE |
| CN115019933B (en) * | 2022-06-16 | 2024-06-21 | 浙江工业大学 | Weak vision training scheme recommendation method integrating GMF and CDAE |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110059262B (en) | 2021-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059262B (en) | Project recommendation model construction method and device based on hybrid neural network and project recommendation method | |
| US11893071B2 (en) | Content recommendation method and apparatus, electronic device, and storage medium | |
| Ma et al. | Effects of user-provided photos on hotel review helpfulness: An analytical approach with deep leaning | |
| Ghorbani et al. | ConvLSTMConv network: a deep learning approach for sentiment analysis in cloud computing | |
| CN111581401B (en) | Local citation recommendation system and method based on depth correlation matching | |
| CN111160350B (en) | Portrait segmentation method, model training method, device, medium and electronic equipment | |
| CN114398961A (en) | Visual question-answering method based on multi-mode depth feature fusion and model thereof | |
| WO2023065859A1 (en) | Item recommendation method and apparatus, and storage medium | |
| CN111382361A (en) | Information pushing method and device, storage medium and computer equipment | |
| CN113010702A (en) | Interactive processing method and device for multimedia information, electronic equipment and storage medium | |
| CN111783903B (en) | Text processing method, text model processing method and device and computer equipment | |
| Karasoy et al. | Spam SMS detection for Turkish language with deep text analysis and deep learning methods | |
| Halvardsson et al. | Interpretation of swedish sign language using convolutional neural networks and transfer learning | |
| CN114201516B (en) | User portrait construction method, information recommendation method and related devices | |
| US20230144138A1 (en) | Machine learning algorithm search with symbolic programming | |
| CN112580352A (en) | Keyword extraction method, device and equipment and computer storage medium | |
| CN111625715A (en) | Information extraction method and device, electronic equipment and storage medium | |
| Chaudhuri | Visual and text sentiment analysis through hierarchical deep learning networks | |
| CN113657272B (en) | Micro video classification method and system based on missing data completion | |
| Sun et al. | Rumour detection technology based on the BiGRU_capsule network | |
| Shukla et al. | Role of hybrid optimization in improving performance of sentiment classification system | |
| CN116932788A (en) | Cover image extraction method, device, equipment and computer storage medium | |
| Baek et al. | Captioning model based on meta-learning using prior-convergence knowledge for explainable images | |
| Ahammed et al. | Meta-transfer Learning for Contextual Emotion Detection in Face Affirmation | |
| Wang et al. | Multi‐Task and Attention Collaborative Network for Facial Emotion Recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210702 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |