CN112699310A - Cold start cross-domain hybrid recommendation method and system based on deep neural network - Google Patents
Cold start cross-domain hybrid recommendation method and system based on deep neural network Download PDFInfo
- Publication number
- CN112699310A CN112699310A CN202011605125.9A CN202011605125A CN112699310A CN 112699310 A CN112699310 A CN 112699310A CN 202011605125 A CN202011605125 A CN 202011605125A CN 112699310 A CN112699310 A CN 112699310A
- Authority
- CN
- China
- Prior art keywords
- data
- user
- cross
- recommendation
- domain
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Finance (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Accounting & Taxation (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Marketing (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a cold start cross-domain hybrid recommendation method and system based on a deep neural network, and belongs to the technical field of network application. The method comprises the following steps: taking the cross user data as training data; generating a set for the text vector; constructing a self-encoder to train text vectors in the set, and outputting a user scoring feature set; the user scoring characteristics in the set are connected in series to serve as the input of the DNN, the predicted scoring data of the user target domain serves as the output, and a cross-domain recommendation model is constructed; the method comprises the steps of obtaining user footprint data of a preset amount, using the user footprint data of the preset amount as recommendation data, inputting the recommendation data into a cross-domain recommendation model for operation, and outputting recommended articles for a user. The method provides a recommendation basis for the target field by learning the correlation nonlinear mapping from the source field to the target field.
Description
Technical Field
The invention relates to the technical field of network application, in particular to a cold start cross-domain hybrid recommendation method and system based on a deep neural network.
Background
With the rapid development of the internet, data also grows exponentially, and in the face of massive data, users are difficult to pick out interesting data from many choices. To improve the user experience, recommendation systems are widely used in different scenarios, such as online shopping, music recommendation, movie recommendation, etc.
The recommendation algorithm is a core part of a recommendation system, and currently, there are mainly three methods, namely Collaborative Filtering (CF) -based recommendation, content-based recommendation, and a hybrid recommendation method. Wherein the CF-based recommendation method uses user and item interaction data, such as user scores. The content-based method finds the relevance between the item and the content according to the metadata of the item or the content, and then recommends similar items to the user based on the past preference records of the user. At present, the CF-based method is widely applied, because the CF-based method can capture the most intuitive evaluation of a user on an article and can be easily expanded into a plurality of scenes, and the content-based method needs to expend certain efforts in different fields to construct a proper feature set to be applied to recommendation. However, the CF-based approach currently has two problems:
1) data are sparse, namely a user score matrix contains a large number of null values;
2) cold start, lack of sufficient historical rating data for the new user. In order to solve the two problems, a hybrid recommendation method is proposed, and the basic idea is to combine the similarity of the user interaction level and the similarity of the item content level, so that a learner proposes to add a lot of additional information to make auxiliary recommendations on the basis of the user rating data, such as basic attributes of the item, the social network of the user, comment data of the user on the item, and the like.
However, in the real world, users are generally reluctant to reveal too much personal information, such as social information, when using new products, and the amount of comment data of users is small when the new products are just used. There is a broad interest in the idea of cross-domain recommendation (cross-domain) that is proposed by scholars without collecting enough user footprint data, and its basic idea is that given two related domains, such as movies and books, users have enough historical footprint data in the source domain, and less historical data in the target domain, then for the target domain, these users can be called cold-start users, because the knowledge of the two domains is relevant, and meaningful recommendations can be provided for the target domain according to the feedback of the users in the source domain.
Disclosure of Invention
Aiming at the problems, the invention provides a cold start cross-domain hybrid recommendation method based on a deep neural network, which comprises the following steps:
acquiring data of a source field and data of a target field, screening the data of the source field and the data of the target field, acquiring cross user data of the source field and the data of the target field, and taking the cross user data as training data;
acquiring basic information and comment data of cross users in training data, performing text vector conversion on the basic information and comment data to generate a text vector, and generating a set aiming at the text vector;
constructing a self-encoder to train text vectors in the set, and outputting a user scoring feature set;
the user scoring characteristics in the set are connected in series to serve as the input of the DNN, the predicted scoring data of the user target domain serves as the output, and a cross-domain recommendation model is constructed;
the method comprises the steps of obtaining user footprint data of a preset amount, using the user footprint data of the preset amount as recommendation data, inputting the recommendation data into a cross-domain recommendation model for operation, and outputting recommended articles for a user.
Optionally, the cross user data is distributed in a preset proportion, and one part of the cross user data is used as training data and the other part of the cross user data is used as test data;
the test data is used to test the accuracy of the cross-domain recommendation model.
Optionally, the method further comprises: constructing an initial model and layering the initial model, wherein the method comprises the following steps:
the input layer acquires data of a source field and data of a target field, screens the data of the source field and the data of the target field, acquires cross user data of the source field and the data of the target field, and takes the cross user data as training data;
the vector representation layer acquires basic information and comment data of cross users in the training data, performs text vector conversion on the basic information and comment data to generate a text vector, and generates a set aiming at the text vector;
the network layer is used for constructing a self-encoder to train the text vectors in the set and outputting a user scoring feature set;
and the output layer is used for serially connecting the user scoring characteristics in the set and used as the input of the DNN network, and the predicted scoring data of the user target domain is used as the output to construct a cross-domain recommendation model.
Optionally, inputting the recommendation data into a cross-domain recommendation model operation, including:
switching the cross-domain recommendation model into a CF-based and content-based mixed recommendation mode, and performing weighted combination on the CF-based and the content-based; and recommending articles for the user according to the recommendation data after the weighted combination is completed.
Optionally, recommending an item for a user includes: recommending according to the user information in the recommendation data, recommending according to the article information in the recommendation data and recommending according to a cross-domain recommendation model.
The invention also provides a system for cold start cross-domain hybrid recommendation based on the deep neural network, which comprises the following steps:
the input layer is used for acquiring data of a source field and data of a target field, screening the data of the source field and the data of the target field, acquiring cross user data of the source field and the data of the target field, and taking the cross user data as training data;
the text vector layer is used for acquiring basic information and comment data of cross users in the training data, converting the text vectors of the basic information and the comment data to generate text vectors and generating a set aiming at the text vectors;
the network layer is used for constructing a self-encoder to train the text vectors in the set and outputting a user scoring feature set;
the output layer is used for serially connecting the user scoring characteristics in the set and taking the user scoring characteristics as input of the DNN network, and the predicted scoring data of the user target domain as output to construct a cross-domain recommendation model;
and the recommendation layer is used for acquiring a preset amount of user footprint data, using the preset amount of user footprint data as recommendation data, inputting the recommendation data into a cross-domain recommendation model for operation, and outputting recommended articles for the user.
Optionally, the cross user data is distributed in a preset proportion, and one part of the cross user data is used as training data and the other part of the cross user data is used as test data;
the test data is used to test the accuracy of the cross-domain recommendation model.
Optionally, inputting the recommendation data into a cross-domain recommendation model operation, including:
switching the cross-domain recommendation model into a CF-based and content-based mixed recommendation mode, and performing weighted combination on the CF-based and the content-based; and recommending articles for the user according to the recommendation data after the weighted combination is completed.
Optionally, recommending an item for a user includes: recommending according to the user information in the recommendation data, recommending according to the article information in the recommendation data and recommending according to a cross-domain recommendation model.
The method provides a recommendation basis for the target field by learning the relevance nonlinear mapping from the source field to the target field, provides a strategy of multi-model hybrid recommendation, mainly based on cross-field recommendation when the user historical data is less, accumulates a certain amount of user historical data in the later period, and can adopt collaborative filtering and a weighting model based on content recommendation to further improve the recommendation effect.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a schematic diagram of a DNN-based cold start cross-domain hybrid recommendation scheme of the present invention;
fig. 3 is a block diagram of the system of the present invention.
Detailed Description
The exemplary embodiments of the present invention will now be described with reference to the accompanying drawings, however, the present invention may be embodied in many different forms and is not limited to the embodiments described herein, which are provided for complete and complete disclosure of the present invention and to fully convey the scope of the present invention to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the accompanying drawings is not intended to be limiting of the invention. In the drawings, the same units/elements are denoted by the same reference numerals.
Unless otherwise defined, terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Further, it will be understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense.
The invention provides a cold start cross-domain hybrid recommendation method based on a deep neural network, which comprises the following steps:
acquiring data of a source field and data of a target field, screening the data of the source field and the data of the target field, acquiring cross user data of the source field and the data of the target field, and taking the cross user data as training data;
acquiring basic information and comment data of cross users in training data, performing text vector conversion on the basic information and comment data to generate a text vector, and generating a set aiming at the text vector;
constructing a self-encoder to train text vectors in the set, and outputting a user scoring feature set;
the user scoring characteristics in the set are connected in series to serve as the input of the DNN, the predicted scoring data of the user target domain serves as the output, and a cross-domain recommendation model is constructed;
the method comprises the steps of obtaining user footprint data of a preset amount, using the user footprint data of the preset amount as recommendation data, inputting the recommendation data into a cross-domain recommendation model for operation, and outputting recommended articles for a user.
The method comprises the following steps that cross user data are distributed according to a preset proportion, wherein one part of the cross user data is used as training data, and the other part of the cross user data is used as test data;
the test data is used to test the accuracy of the cross-domain recommendation model.
Wherein, the method further comprises: constructing an initial model and layering the initial model, wherein the method comprises the following steps:
the input layer acquires data of a source field and data of a target field, screens the data of the source field and the data of the target field, acquires cross user data of the source field and the data of the target field, and takes the cross user data as training data;
the vector representation layer acquires basic information and comment data of cross users in the training data, performs text vector conversion on the basic information and comment data to generate a text vector, and generates a set aiming at the text vector;
the network layer is used for constructing a self-encoder to train the text vectors in the set and outputting a user scoring feature set;
and the output layer is used for serially connecting the user scoring characteristics in the set and used as the input of the DNN network, and the predicted scoring data of the user target domain is used as the output to construct a cross-domain recommendation model.
Inputting recommendation data into cross-domain recommendation model operation, wherein the cross-domain recommendation model operation comprises the following steps:
switching the cross-domain recommendation model into a CF-based and content-based mixed recommendation mode, and performing weighted combination on the CF-based and the content-based; and recommending articles for the user according to the recommendation data after the weighted combination is completed.
Wherein, recommending articles for users comprises: recommending according to the user information in the recommendation data, recommending according to the article information in the recommendation data and recommending according to a cross-domain recommendation model.
The invention is further illustrated by the following examples:
assuming that the source domain and target domain knowledge have relevance, such as movies and books, the hierarchy of cross-domain recommendations, as shown in fig. 2, includes an input layer, a vector representation layer, a network layer, and an output layer, respectively.
An input layer;
and processing the data of the source domain and the target domain, and screening out cross users as training and testing data of the network. The input layer contains collected data of the source field, and can comprise basic information of the user (such as age, gender and the like), comment data of the user on the item and grading data of the user.
A vector representation layer;
for basic information and comment data of users, a word2vec model pre-trained on a large public data set can be used for text vector representation because the basic information and comment data are mainly in a text form. The user rating data may be represented in the form of a matrix as in fig. 2, with each row representing the rating data for each item by the user. Because the number of the items is larger, and the number of the items which are scored by the user is relatively small, a large number of null values exist in the matrix.
TABLE 2
Because the high-latitude sparse characteristic is not suitable for training of multilayer complex neural networks, the network training parameter quantity is huge and convergence is difficult. Therefore, the dimensionality reduction is needed, the self-encoder model is adopted in the method, the low latitude data are encoded through a multilayer neural network, and the high latitude data are reconstructed, and practice proves that the algorithm is more effective than the traditional dimensionality reduction methods such as PCA and LDA.
Two parts are required to construct a self-encoder: an encoder and a decoder. The encoder compresses the input vector into a latent semantic space representation, which may be represented by the function f (x), and the decoder reconstructs the latent semantic space representation into output, which may be represented by the function g (x), where f (x) and g (x) are both neural network models. The loss function is the square error loss, and the network is trained by using a gradient descent method.
A network layer;
and directly connecting the user basic information and comment feature set represented by using word2vec with the user scoring feature set represented by the self-encoder in series to serve as the input of the DNN.
The DNN is a hierarchical network, divided into an input layer, a hidden layer and an output layer. The number of the hidden layers can be designed according to the size of data volume, the data volume is large generally, the number of layers of the hidden layers can be increased, a network of each layer is composed of a plurality of perceptrons, a sigmoid or ReLU is adopted as an activation function, cross entropy loss is adopted as a loss function, and the network is trained by using a gradient descent method.
An output layer;
the cross-domain recommendation model inputs user data in a source field and outputs the user data in a prediction scoring data corresponding to a target field of a user, so that reasonable recommendation is made for the target field according to knowledge in the source field.
Collaborative filtering and weighted combination based on content recommendations
After a certain amount of user footprint data is accumulated, the cross-domain recommendation model can be switched to the CF-based and content-based recommendation methods which are widely applied at present. The patent adopts a recommendation method of the weighted combination of CF-based and content-based, and the defects can be mutually compensated by using the weighted combination.
As for the CF-based recommendation algorithm, the basic idea is to recommend commodities to users according to the previous preference of the users and the selection of the users with similar interests. The calculation of the similarity is performed by collecting the user-item scores (m x n matrix). The method specifically comprises the following steps:
a user-based (user-based) recommendation method;
the main consideration is that as long as finding out articles similar to the user's liking and predicting the target user's score for the corresponding articles, several articles with the highest scores can be found and recommended to the user.
Item-based recommendation methods;
similar to user-based collaborative filtering, but for similarity between items
And if the scores of the target users for some items are found, recommending a plurality of similar items with the highest scores to the users.
Model-based (model-based) recommendation methods;
there are many model-based methods, which are hot spots of current academic and industrial research
The method is realized by modeling by using the idea of machine learning, and is generally applicable to the background of a large amount of data, such as a common associated algorithm, a clustering algorithm, classification and regression, matrix decomposition and the like.
The three recommendation algorithms contained in the above three methods for CF-based, the user-based method has higher computational complexity than item-based, and can help users find new types of articles with surprise. Whereas collaborative filtering based on items may be computed off-line since the similarity of the considered items does not change over time, but the diversity of recommended items is less. In practical use, for small recommendation systems, item-based usage is the mainstream, while for large recommendation systems, user-based or model-based usage may be employed.
Because the conventional recommendation system mainly adopts a collaborative filtering method and a content-based recommendation method, but both methods need to recommend according to historical favorite data of a user, and the problem of cold start generally exists for new online products. In addition, the comprehensive requirements of the user cannot be met by adopting a single recommendation algorithm, the patent provides a strategy of multi-model mixed recommendation, when the user historical data are less, a certain amount of user historical data are accumulated in the later period mainly according to cross-domain recommendation, and the recommendation effect can be further improved by adopting collaborative filtering and a weighting model based on content recommendation.
The invention further provides a system 200 for cold start cross-domain hybrid recommendation based on a deep neural network, as shown in fig. 3, including:
the input layer 201 is used for acquiring data of a source field and a target field, screening the data of the source field and the target field, acquiring cross user data of the source field and the target field, and using the cross user data as training data;
the text vector layer 202 is used for acquiring basic information and comment data of cross users in the training data, performing text vector conversion on the basic information and comment data to generate a text vector, and generating a set aiming at the text vector;
the network layer 203 is used for constructing a self-encoder to train the text vectors in the set and outputting a user scoring feature set;
the output layer 204 is used for serially connecting the user scoring characteristics in the set and taking the user scoring characteristics as input of the DNN network, and the predicted scoring data of the user target domain as output to construct a cross-domain recommendation model;
and the recommendation layer 205 is used for acquiring a preset amount of user footprint data, using the preset amount of user footprint data as recommendation data, inputting the recommendation data into a cross-domain recommendation model for operation, and outputting recommended articles for the user.
The method comprises the following steps that cross user data are distributed according to a preset proportion, wherein one part of the cross user data is used as training data, and the other part of the cross user data is used as test data;
the test data is used to test the accuracy of the cross-domain recommendation model.
Inputting recommendation data into cross-domain recommendation model operation, wherein the cross-domain recommendation model operation comprises the following steps:
switching the cross-domain recommendation model into a CF-based and content-based mixed recommendation mode, and performing weighted combination on the CF-based and the content-based; and recommending articles for the user according to the recommendation data after the weighted combination is completed.
Wherein, recommending articles for users comprises: recommending according to the user information in the recommendation data, recommending according to the article information in the recommendation data and recommending according to a cross-domain recommendation model.
The method provides a recommendation basis for the target field by learning the relevance nonlinear mapping from the source field to the target field, provides a strategy of multi-model hybrid recommendation, mainly based on cross-field recommendation when the user historical data is less, accumulates a certain amount of user historical data in the later period, and can adopt collaborative filtering and a weighting model based on content recommendation to further improve the recommendation effect.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application 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 scheme in the embodiment of the application can be implemented by adopting various computer languages, such as object-oriented programming language Java and transliterated scripting language JavaScript.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. 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.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While the preferred embodiments of the present application 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 alterations and modifications as fall within the scope of the application.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the spirit and scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.
Claims (9)
1. A method of cold-start cross-domain hybrid recommendation based on a deep neural network, the method comprising:
acquiring data of a source field and data of a target field, screening the data of the source field and the data of the target field, acquiring cross user data of the source field and the data of the target field, and taking the cross user data as training data;
acquiring basic information and comment data of cross users in training data, performing text vector conversion on the basic information and comment data to generate a text vector, and generating a set aiming at the text vector;
constructing a self-encoder to train text vectors in the set, and outputting a user scoring feature set;
the user scoring characteristics in the set are connected in series to serve as the input of the DNN, the predicted scoring data of the user target domain serves as the output, and a cross-domain recommendation model is constructed;
the method comprises the steps of obtaining user footprint data of a preset amount, using the user footprint data of the preset amount as recommendation data, inputting the recommendation data into a cross-domain recommendation model for operation, and outputting recommended articles for a user.
2. The method of claim 1, wherein the cross user data is distributed in a predetermined ratio, one part being training data and one part being test data;
the test data is used for testing the accuracy of the cross-domain recommendation model.
3. The method of claim 1, further comprising: constructing an initial model and layering the initial model, wherein the method comprises the following steps:
the input layer acquires data of a source field and data of a target field, screens the data of the source field and the data of the target field, acquires cross user data of the source field and the data of the target field, and takes the cross user data as training data;
the vector representation layer acquires basic information and comment data of cross users in the training data, performs text vector conversion on the basic information and comment data to generate a text vector, and generates a set aiming at the text vector;
the network layer is used for constructing a self-encoder to train the text vectors in the set and outputting a user scoring feature set;
and the output layer is used for serially connecting the user scoring characteristics in the set and used as the input of the DNN network, and the predicted scoring data of the user target domain is used as the output to construct a cross-domain recommendation model.
4. The method of claim 1, the inputting recommendation data into a cross-domain recommendation model operation, comprising:
switching the cross-domain recommendation model into a CF-based and content-based mixed recommendation mode, and performing weighted combination on the CF-based and the content-based; and recommending articles for the user according to the recommendation data after the weighted combination is completed.
5. The method of claim 4, the recommending items for a user, comprising: recommending according to the user information in the recommendation data, recommending according to the article information in the recommendation data and recommending according to a cross-domain recommendation model.
6. A system for cold-start cross-domain hybrid recommendation based on a deep neural network, the system comprising:
the input layer is used for acquiring data of a source field and data of a target field, screening the data of the source field and the data of the target field, acquiring cross user data of the source field and the data of the target field, and taking the cross user data as training data;
the text vector layer is used for acquiring basic information and comment data of cross users in the training data, converting the text vectors of the basic information and the comment data to generate text vectors and generating a set aiming at the text vectors;
the network layer is used for constructing a self-encoder to train the text vectors in the set and outputting a user scoring feature set;
the output layer is used for serially connecting the user scoring characteristics in the set and taking the user scoring characteristics as input of the DNN network, and the predicted scoring data of the user target domain as output to construct a cross-domain recommendation model;
and the recommendation layer is used for acquiring a preset amount of user footprint data, using the preset amount of user footprint data as recommendation data, inputting the recommendation data into a cross-domain recommendation model for operation, and outputting recommended articles for the user.
7. The system of claim 6, wherein the cross user data is distributed in a predetermined ratio, one part being training data and one part being test data;
the test data is used for testing the accuracy of the cross-domain recommendation model.
8. The system of claim 6, the inputting recommendation data into a cross-domain recommendation model operation, comprising:
switching the cross-domain recommendation model into a CF-based and content-based mixed recommendation mode, and performing weighted combination on the CF-based and the content-based; and recommending articles for the user according to the recommendation data after the weighted combination is completed.
9. The system of claim 8, the recommending items for a user, comprising: recommending according to the user information in the recommendation data, recommending according to the article information in the recommendation data and recommending according to a cross-domain recommendation model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011605125.9A CN112699310A (en) | 2020-12-30 | 2020-12-30 | Cold start cross-domain hybrid recommendation method and system based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011605125.9A CN112699310A (en) | 2020-12-30 | 2020-12-30 | Cold start cross-domain hybrid recommendation method and system based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112699310A true CN112699310A (en) | 2021-04-23 |
Family
ID=75512343
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011605125.9A Pending CN112699310A (en) | 2020-12-30 | 2020-12-30 | Cold start cross-domain hybrid recommendation method and system based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112699310A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN113435984A (en) * | 2021-08-27 | 2021-09-24 | 苏州浪潮智能科技有限公司 | Cross-domain recommendation method, system, storage medium and equipment |
| CN113935477A (en) * | 2021-12-17 | 2022-01-14 | 深圳佑驾创新科技有限公司 | Recommendation model training method, recommendation method and computer-readable storage medium |
| WO2023173550A1 (en) * | 2022-03-14 | 2023-09-21 | 平安科技（深圳）有限公司 | Cross-domain data recommendation method and apparatus, and computer device and medium |
-
2020
- 2020-12-30 CN CN202011605125.9A patent/CN112699310A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
| CN113435984A (en) * | 2021-08-27 | 2021-09-24 | 苏州浪潮智能科技有限公司 | Cross-domain recommendation method, system, storage medium and equipment |
| CN113935477A (en) * | 2021-12-17 | 2022-01-14 | 深圳佑驾创新科技有限公司 | Recommendation model training method, recommendation method and computer-readable storage medium |
| CN113935477B (en) * | 2021-12-17 | 2022-02-22 | 深圳佑驾创新科技有限公司 | Recommendation model training method, recommendation method and computer-readable storage medium |
| WO2023173550A1 (en) * | 2022-03-14 | 2023-09-21 | 平安科技（深圳）有限公司 | Cross-domain data recommendation method and apparatus, and computer device and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111241311B (en) | Media information recommendation method and device, electronic equipment and storage medium | |
| CN111931062B (en) | Training method and related device of information recommendation model | |
| CN111310063B (en) | Neural network-based article recommendation method for memory perception gated factorization machine | |
| CN110619081B (en) | News pushing method based on interactive graph neural network | |
| CN112699310A (en) | Cold start cross-domain hybrid recommendation method and system based on deep neural network | |
| CN112800334A (en) | Collaborative filtering recommendation method and device based on knowledge graph and deep learning | |
| CN111241394B (en) | Data processing method, data processing device, computer readable storage medium and electronic equipment | |
| Wen et al. | Neural attention model for recommendation based on factorization machines | |
| CN111949886B (en) | Sample data generation method and related device for information recommendation | |
| CN109902201A (en) | A kind of recommended method based on CNN and BP neural network | |
| CN112016002A (en) | Mixed recommendation method integrating comment text level attention and time factors | |
| US20230237093A1 (en) | Video recommender system by knowledge based multi-modal graph neural networks | |
| Zan et al. | UDA: A user-difference attention for group recommendation | |
| Khoali et al. | Advanced recommendation systems through deep learning | |
| Alfarhood et al. | DeepHCF: a deep learning based hybrid collaborative filtering approach for recommendation systems | |
| JP2023024932A (en) | System for multi-modal transformer-based item categorization, data processing system, data processing method, and computer implemented method | |
| Zhang et al. | Deep learning for recommender systems | |
| US20240037133A1 (en) | Method and apparatus for recommending cold start object, computer device, and storage medium | |
| Madani et al. | A review-based context-aware recommender systems: Using custom ner and factorization machines | |
| Zeng | Application of conditional random field model based on machine learning in online and offline integrated educational resource recommendation | |
| George et al. | PERKC: Personalized kNN With CPT for Course Recommendations in Higher Education | |
| Bang et al. | Collective matrix factorization using tag embedding for effective recommender system | |
| Hien et al. | A Deep Learning Model for Context Understanding in Recommendation Systems | |
| Kanjirathinkal | Explainable Recommendations | |
| Xu et al. | Graph Neural Network for User-Item Recommendation |
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 |