CN114996566A - Intelligent recommendation system and method for industrial internet platform - Google Patents
Intelligent recommendation system and method for industrial internet platform Download PDFInfo
- Publication number
- CN114996566A CN114996566A CN202210439549.5A CN202210439549A CN114996566A CN 114996566 A CN114996566 A CN 114996566A CN 202210439549 A CN202210439549 A CN 202210439549A CN 114996566 A CN114996566 A CN 114996566A
- Authority
- CN
- China
- Prior art keywords
- information
- user
- vector
- feature
- management module
- 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
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- 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
- G06F16/906—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/12—Use of codes for handling textual entities
- G06F40/126—Character encoding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the technical field of computer recommendation, and discloses an intelligent recommendation system and method for an industrial internet platform, wherein the system comprises a background management module, a user side, a recommendation management module and a front-end display module; the method comprises the following steps: (1) carrying out data preprocessing on the information of the platform in the background management module; (2) carrying out data preprocessing on user information; (3) text-topic feature extraction on platform text information; (4) related enterpriseExtracting user-interest characteristics of business user information; (5) the vectors y obtained in the procedures (3) and (4) CNN 、y AFM Through the fusion of low-order and high-order feature interaction, probability output of the predicted love degree is obtained through sigmoid nonlinear conversion, and then the recommended information is sequenced according to the probability of the love degree to obtain topk to be recommended; (6) and transmitting the topk to be recommended to a front-end display module to form a recommendation list and recommending the recommendation list to the corresponding enterprise user.
Description
Technical Field
The invention relates to the technical field of computer recommendation, in particular to an intelligent recommendation system and method for an industrial internet platform.
Background
In recent years, with the rapid development of the internet era, the amount of information on the internet has greatly exploded, which leads us to be in an era of data explosion. Because of the coming of this age, when browsing information on a certain platform, there is a problem of information overload, that is, a user cannot obtain the part of information really valuable to himself from the excessive information, and the use efficiency of the information is extremely low. In addition, with the rise of the industrial internet in recent years, more and more industrial internet platforms gradually emerge, and the platforms aim to provide various product and service transaction platforms for enterprise users, and simultaneously can provide various latest relevant news and various policies for the enterprise users, so as to help the enterprise users to know the latest dynamic and supply and demand information of the industry in time. At present, small-sized industrial internet platforms created by a plurality of human enterprises are good in the market besides the well-known Halkaos industrial internet platform, the Xuhan cloud industrial internet platform and the like.
For the appearance of numerous industrial internet platforms, it takes a lot of time and effort and is extremely inefficient for enterprise users browsing access platforms to find products, policy information and the like meeting their own needs. In recent years, part of recommendations about industrial internet platforms are based on a traditional recommendation algorithm based on collaborative filtering, but the traditional recommendation algorithm cannot solve the current practical problems such as data sparseness and cold start. In an industrial internet platform, the importance of a certain product or information to different enterprise users is different, so that the above pain problems such as data sparseness and cold start cannot be solved only by using a traditional recommendation algorithm.
Compared with other information website platforms, the information and the user of the industrial internet platform have the characteristics of being more vivid: such as the relevance of the diversity and hidden features of news information and various industry solutions in the platform, and the nature of users in the platform, although the number is not as large as the order of magnitude of platform information, each user has relatively personalized features and preferences (industry dynamics, solutions, etc.). Aiming at the characteristics of the industrial internet platform, the patent provides a new recommendation model which relates to a convolutional neural network, an attention mechanism and a factorization machine and faces the industrial internet platform. Patent document CN109241424A discloses a recommendation method, and the proposed model uses three technologies, namely convolutional neural network, attention mechanism and factorization machine, but the whole process of processing data and the model are not well suitable for the industrial internet platform. The invention is innovative according to the unique characteristics of the industrial Internet platform.
Disclosure of Invention
The invention provides an intelligent recommendation system facing an industrial internet platform and a working method thereof, aiming at the problems, firstly, the problem of information overload in the industrial internet platform is solved, secondly, the cold start problem of a new enterprise user in recommendation is solved, and the problem of inconsistent importance weight of information such as a certain product, an industry solution and the like to the user is solved.
The technical scheme for solving the technical problem of the invention is as follows:
an intelligent recommendation system facing an industrial Internet platform comprises a background management module, a user side, a recommendation management module and a front-end display module,
the background management module is used for carrying out feature classification on the information data of the platform, and carrying out addition, deletion, modification, check and storage management on the information data; the information data mainly comprises an industry news policy, an industry solution, various products, an expert database and the like on an industry internet platform.
The user side comprises a user information management module and a user behavior record management module, wherein the user information management module is responsible for registering new users of enterprise users and storing user registration information, and the user behavior record management module is responsible for recording browsing record logs of the users on the platform; such as news policy browsing, industry solutions browsing, platform retrieval records, etc.
The front-end display module is mainly used for displaying a list to be recommended which is most matched with the front-end display module for enterprise users;
the recommendation management module takes information data of the background management module and information data of a user side as input, performs feature modeling on the information data of the background management module through a convolutional neural network to obtain a text-theme feature model, performs feature modeling on the information data of the user side through an FM (frequency modulation) combined with an Attention Mechanism (Attention Mechanism) to obtain a user-interest feature model, performs feature interaction fusion on feature vectors of the text-theme feature model and the user-interest feature model, performs sigmoid nonlinear conversion to obtain a probability value of the predicted liking degree of the recommended information, sorts the recommended information according to the size of the probability value of the liking degree, and obtains k recommended results topk as output to be transmitted to the front-end display module.
A working method of an intelligent recommendation system facing an industrial Internet platform comprises the following processes:
(1) carrying out data preprocessing on the information of the platform in the background management module: carrying out jieba word segmentation on the text data of the platform to convert the text data into a plurality of token units, and obtaining a vector X with a fixed length by a plurality of tokens obtained by word segmentation in a word2vector mode i ；
(2) Carrying out data preprocessing on user information: one-hot coding is carried out on the registration information and the behavior record data information of the user, and the registration information and the behavior record data information are converted into word vectors Q; this encoding format uses an n 'bit state register to encode n' states, each having its own independent register bit and only one bit being active at any one time. Is the most common way of binary vector representation. For example, "china, usa, france" is "100, 010, 001" by one-hot coding, respectively. All data about the user is converted into a vector representation by this coding.
(3) Text-to-subject feature with respect to platform textual informationExtraction: the vector X obtained in the procedure (1) i The embedding input of the convolutional neural network part serving as the recommendation management module is transmitted into a hidden layer containing convolutional layers for convolution calculation, and finally a text-subject feature vector y is output CNN (ii) a The convolutional neural network may be any convolutional neural network, such as a prior art m-layer Convolutional Neural Network (CNN).
(4) User-interest feature extraction on enterprise user information: the vector Q obtained in the flow (2) is taken as another part of a recommendation management module and introduced into an input part of a Factorization Machine (FM) of the attention mechanism network, and the part is used for obtaining a user-interest feature vector y with each user having a unique information weight through the Factorization Machine (FM) and attention mechanism calculation AFM The formula is as follows:
in formula (I), y AFM For the user-interest feature vector, w 0 Is a global constant bias term, n is the total number of the enterprise user information user-interest characteristics, i and j are values greater than 0 and less than n, w i For first order eigenvector parameters, p T Is a parameter vector, a i，j Is a characteristic value Q i And Q j Corresponding hidden vector v i And v j Attention score of v i A K-dimensional hidden vector of the user-interest characteristics of the ith enterprise user information; v. of j A K-dimensional hidden vector, Q, of information user-interest features of the jth enterprise user i For the ith feature vector Q, Q in the above-mentioned flow (2) j The j-th feature vector Q in the above flow (2);
the essence of the attention mechanism network is a simple structure of a full connection layer and a softmax output layer, and the mathematical expression of the structure is as follows:
a' i,j ＝h T ReLu(W(v i ⊙v j )Q i Q j +b) (II)
in equations (II), (III), h is the weight vector from the fully-connected layer to the softmax output layer, h T Is a transposed matrix of weight vectors h, ReLu is an activation function of the attention network, W is a weight matrix of the feature cross layer to the attention network full link layer, v i A K-dimensional hidden vector of the i-th enterprise user information user-interest feature; v. of j A K-dimensional hidden vector, Q, of information user-interest features of the jth enterprise user i For the ith feature vector Q, Q in the above-mentioned flow (2) j In the above flow (2), the jth eigenvector Q and b are bias vectors, and the learned model parameters are the weight matrix W from the eigen cross layer to the attention network full link layer, the bias vector b, and the weight vector h, a 'from the full link layer to the softmax output layer' i，j For the calculated attention score prophase data, a i，j Is a' i，j The results are mapped to an attention score between 0-1 by softmax.
(5) The vectors y obtained in the procedures (3) and (4) CNN 、y AFM Through the fusion of low-order and high-order feature interaction, probability output of predicting the love degree is obtained through sigmoid nonlinear conversion, then the recommended information is sequenced according to the probability of the love degree to obtain topk to be recommended, and the formula is as follows:
y＝sigmoid(y CNN +y AFM ) (IV)
in formula (IV), y is a probability value of the preference degree of the recommended information, a value is 0-1, and topk is k results recommended by the system;
(6) and transmitting the topk to be recommended to a front-end display module to form a recommendation list and recommending the recommendation list to the corresponding enterprise user.
The effect provided in the summary of the invention is only the effect of the embodiment, not all the effects of the invention, and one of the above technical solutions has the following advantages or beneficial effects:
the invention separately models the industrial internet platform according to the unique characteristics of the information data and the user information of the industrial internet platform, can extract the characteristics of the platform information more deeply through the convolutional neural network, effectively solves the correlation of the diversity and the hidden characteristics of the platform information, solves the problems of sparseness of the user-theme matrix data and different information weights (different information preference degrees and different information weights because the user has personalized characteristic preference) through the combination of the attention mechanism and a factorization machine, and carries out modeling in different modes according to the characteristics of the user and the information in the industrial internet platform so as to make a recommendation result more humanized.
The problem of cold starting of articles is solved through the CNN in the step 3, the cold starting of the articles refers to how to recommend new articles to users who may be interested in the articles, the articles of the invention refer to useful information in an industrial Internet platform, and the semantic features of word sequences of text information can be effectively captured through the powerful local feature extraction capability of the CNN in the step 3, so that a group of articles with high similarity are obtained and are recommended to the users finally (because the article information has high similarity), and the problem of cold starting of the articles in recommendation is solved;
according to the invention, through the attention mechanism introduced in the step 4, the attention mechanism is thought to enable the model to learn to distribute the attention of the model to highlight different influences of different characteristics on modeling, different information in the platform has different importance on each enterprise user, and the same information has different information amount when different users are modeled, so that the problem of different information importance weights is solved by introducing the attention mechanism.
According to the invention, the factor decomposition machine (FM) combined with the attention mechanism in the step 4 considers the second-order cross of the features, so that each vector is fully learned, and the problem of difficulty in training caused by data sparseness is solved.
Drawings
FIG. 1 is a schematic flow chart of the system of the present invention;
FIG. 2 is a schematic diagram of a system model of the present invention;
Detailed Description
In order to clearly explain the technical features of the present invention, the following detailed description of the present invention is provided with reference to the accompanying drawings. The following disclosure provides many different embodiments, or examples, for implementing different features of the invention. To simplify the disclosure of the present invention, specific example components and arrangements are described below. Furthermore, the present invention may repeat reference numerals and/or letters in the various examples. This repetition is for the purpose of simplicity and clarity and does not in itself dictate a relationship between the various embodiments and/or configurations discussed. It should be noted that the components illustrated in the figures are not necessarily drawn to scale. Descriptions of well-known components and processing techniques and processes are omitted so as to not unnecessarily limit the invention.
An intelligent recommendation system for an industrial Internet platform comprises a background management module, a user side, a recommendation management module and a front-end display module,
the background management module is used for carrying out feature classification on the information data of the platform, and carrying out addition, deletion, modification, check and storage management on the information data; the information data mainly comprises an industry news policy, an industry solution, various products, an expert database and the like on an industry internet platform.
The user side comprises a user information management module and a user behavior record management module, wherein the user information management module is responsible for registering new users of enterprise users and storing user registration information, and the user behavior record management module is responsible for recording browsing record logs of the users on the platform; such as news policy browsing, industry solutions browsing, platform retrieval records, etc.
The front-end display module is mainly used for displaying a list to be recommended which is most matched with the front-end display module for enterprise users;
the recommendation management module takes information data of the background management module and information data of a user side as input, performs feature modeling on the information data of the background management module through a convolutional neural network to obtain a text-theme feature model, performs feature modeling on the information data of the user side through an FM (frequency modulation) combined with an Attention Mechanism (Attention Mechanism) to obtain a user-interest feature model, performs feature interaction fusion on feature vectors of the text-theme feature model and the user-interest feature model, performs sigmoid nonlinear conversion to obtain a predicted probability value of the likeness of the recommended information, sorts the recommended information according to the magnitude of the probability value of the likeness, and obtains k recommended results topk as output to be transmitted to the front-end display module.
An operating method of an intelligent recommendation system facing an industrial internet platform comprises the following steps:
(1) carrying out data preprocessing on the information of the platform in the background management module: carrying out jieba word segmentation on the text data of the platform to convert the text data into n token units, and obtaining a vector X with a fixed length from the n tokens obtained by word2vector mode i ；
(2) Carrying out data preprocessing on user information: one-hot coding is carried out on the registration information and the behavior record data information of the user, and the registration information and the behavior record data information are converted into word vectors Q; this encoding format uses an n 'bit state register to encode n' states, each having its own independent register bit and only one bit being active at any one time. Is the most common way of binary vector representation. For example, "china, usa, france" is "100, 010, 001" respectively by one-hot coding. All data about the user is converted into a vector representation by this coding.
(3) Text-topic feature extraction on platform text information: the vector X obtained in the procedure (1) i The embedding input of the convolutional neural network part serving as the recommendation management module is transmitted into a hidden layer containing convolutional layers for convolution calculation, and finally, a text-subject feature vector y is output CNN (ii) a The convolutional neural network may be any convolutional neural network, such as a prior art m-layer Convolutional Neural Network (CNN).
(4) User-interest features on enterprise user informationExtraction: the vector Q obtained in the flow (2) is taken as another part of a recommendation management module and introduced into an input part of a Factorization Machine (FM) of the attention mechanism network, and the part is used for obtaining a user-interest feature vector y with each user having a unique information weight through the Factorization Machine (FM) and attention mechanism calculation AFM 。
The method is characterized in that FM (factor decomposition) is used as an improved version of a logistic regression model, the problem that model parameters are difficult to train in a recommended sparse data scene (characteristic vectors are formed through one-hot coding) is solved, second-order cross of characteristics is considered, a polynomial is added on the basis of a linear model and used for describing the second-order cross of the characteristics, and the formula of the FM is as follows:
in the formula (V), y FM For the user-interest feature vector, w 0 Is a global constant bias term, n is the total number of user-interest features of the enterprise user information, i and j are values greater than 0 and less than n, w i Is a first order eigenvector parameter, w ij As second-order cross-feature parameters, Q i For the ith feature vector Q, Q in the above-mentioned flow (2) j The j-th feature vector Q in the above flow (2);
the attention mechanism is equivalent to a weighted average, and the value of attention is the weight thereof, so as to judge the importance of interaction between different features, therefore, the improved formula of the invention is as follows:
in formula (I), y AFM For the user-interest feature vector, w 0 Is a global constant bias term, n is the total number of the enterprise user information user-interest characteristics, i and j are values greater than 0 and less than n, w i For first order eigenvector parameters, p T Is a parameter vector, a i，j Is a characteristic value Q i And Q j Corresponding hidden vector v i And v j Attention score of (v) i A K-dimensional hidden vector of the i-th enterprise user information user-interest feature; v. of j A K-dimensional hidden vector, Q, of information user-interest features of the jth enterprise user i For the ith feature vector Q, Q in the above-mentioned flow (2) j The j-th feature vector Q in the above flow (2);
the modified formula differs from the FM formula by a third term, the most important of which is the attention score a i,j The simplest method is to use a weighting parameter to represent, but in order to prevent the weighting parameter from converging difficultly due to sparse cross feature data, an attention network between two feature cross layers and a pooling layer is used to generate the attention score, and the nature of the attention network is a simple structure of a full connection layer and a softmax output layer, and the mathematical expression of the attention network is as follows:
a' i,j ＝h T ReLu(W(v i ⊙v j )Q i Q j +b) (II)
in equations (II), (III), h is the weight vector from the fully-connected layer to the softmax output layer, h T Is a transposed matrix of weight vectors h, ReLu is an activation function of the attention network, W is a weight matrix of the feature cross layer to the attention network full link layer, v i A K-dimensional hidden vector of the i-th enterprise user information user-interest feature; v. of j A K-dimensional hidden vector, Q, of information user-interest features of the jth enterprise user i For the ith feature vector Q, Q in the above-mentioned flow (2) j In the above flow (2), the jth eigenvector Q and b are bias vectors, and the learned model parameters are the weight matrix W from the eigen cross layer to the attention network full link layer, the bias vector b, and the weight vector h, a 'from the full link layer to the softmax output layer' i，j For the calculated attention score prophase data, a i，j Is a' i，j Will pass through softmaxThe results map to an attention score between 0-1.
(5) The vectors y obtained in the processes (3) and (4) CNN 、y AFM Through the fusion of low-order and high-order feature interaction, probability output of predicting the love degree is obtained through sigmoid nonlinear conversion, then the recommended information is sequenced according to the probability of the love degree to obtain topk to be recommended, and the formula is as follows:
y＝sigmoid(y CNN +y AFM ) (IV)
in formula (IV), y is a probability value of the preference degree of the recommended information, a value is 0-1, and topk is k results recommended by the system;
(6) and transmitting the topk to be recommended to a front-end display module to form a recommendation list and recommending the recommendation list to the corresponding enterprise user.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, the scope of the present invention is not limited thereto, and various modifications and variations which do not require inventive efforts and which are made by those skilled in the art are within the scope of the present invention.
Claims (3)
1. A recommendation system of an industrial Internet platform based on a convolutional neural network and an attention mechanism is characterized by comprising a background management module, a user side, a recommendation management module and a front-end display module;
the background management module is used for carrying out feature classification on the information data of the platform, and carrying out addition, deletion, modification, check and storage management on the information data;
the user side comprises a user information management module and a user behavior record management module, the user information management module is responsible for registering new users of enterprise users and storing user registration information, and the user behavior record management module is responsible for recording browsing record logs of the users on the platform;
the recommendation management module is used for inputting information data of the background management module and information data of the user side, performing feature modeling on the information data of the background management module through a convolutional neural network to obtain a text-theme feature model, performing feature modeling on the information data of the user side through a factorization machine in combination with an attention mechanism to obtain a user-interest feature model, performing feature interactive fusion on feature vectors emidding of the text-theme feature model and the user-interest feature model, performing sigmoid nonlinear conversion to obtain a probability value of the predicted likeness of the recommended information, sequencing the recommended information according to the probability value of the likeness to obtain k recommended results topk, and outputting the k recommended results topk to the front-end display module;
the front-end display module displays a list to be recommended which is most matched with the front-end display module for the enterprise user.
2. A working method of a recommendation system of an industrial Internet platform based on a convolutional neural network and an attention mechanism is characterized by comprising the following processes:
(1) carrying out data preprocessing on the information of the platform in the background management module: carrying out jieba word segmentation on the text data of the platform to convert the text data into a plurality of token units, and obtaining a vector X with a fixed length by a plurality of tokens obtained by word segmentation in a word2vector mode i ；
(2) Carrying out data preprocessing on user information: performing one-hot coding on the registration information and behavior record data information of the user, and converting the registration information and the behavior record data information into word vectors Q;
(3) text-topic feature extraction on platform text information: the vector X obtained in the procedure (1) i The embedding input of the convolutional neural network part serving as the recommendation management module is transmitted into a hidden layer containing convolutional layers for convolution calculation, and finally a text-subject feature vector y is output CNN ；
(4) User-interest feature extraction on enterprise user information: and (3) introducing the vector Q obtained in the flow (2) as another part of the recommendation management module into an input part of a factorization machine of the attention mechanism network, wherein the formula is as follows:
in formula (I), y AFM For the user-interest feature vector, w 0 Is a global constant bias term, n is the total number of the enterprise user information user-interest characteristics, i and j are values greater than 0 and less than n, w i For first order eigenvector parameters, p T Is a parameter vector, a i，j Is a characteristic value Q i And Q j Corresponding hidden vector v i And v j Attention score of v i A K-dimensional hidden vector of the user-interest characteristics of the ith enterprise user information; v. of j A K-dimensional hidden vector, Q, of information user-interest features of the jth enterprise user i For the ith feature vector Q, Q in the above-mentioned flow (2) j The j-th feature vector Q in the above flow (2);
a' i,j ＝h T ReLu(W(v i ⊙v j )Q i Q j +b) (II)
in equations (II), (III), h is the weight vector of the fully-connected layer to the softmax output layer, h T Is a transposed matrix of weight vectors h, ReLu is an activation function of the attention network, W is a weight matrix of the feature cross layer to the attention network full link layer, v i A K-dimensional hidden vector of the i-th enterprise user information user-interest feature; v. of j A K-dimensional hidden vector, Q, of information user-interest features of the jth enterprise user i For the ith feature vector Q, Q in the above-mentioned flow (2) j In the above flow (2), the jth eigenvector Q, b is a bias vector, and the learned model parameters are the weight matrix W from the eigen cross layer to the attention network fully-connected layer, the bias vector b, and the weight vector h, a 'from the fully-connected layer to the softmax output layer' i，j For the calculated attention score prophase data, a i，j Is a' i，j Mapping the results to an attention score between 0-1 by softmax;
(5) The vectors y obtained in the processes (3) and (4) CNN 、y AFM Through the fusion of low-order and high-order feature interaction, probability output of predicting the love degree is obtained through sigmoid nonlinear conversion, then the recommended information is sequenced according to the probability of the love degree to obtain topk to be recommended, and the formula is as follows:
y＝sigmoid(y CNN +y AFM ) (IV)
in formula (IV), y is a probability value of the preference degree of the recommended information, a value is 0-1, and topk is k results recommended by the system;
(6) and transmitting the topk to be recommended to a front-end display module to form a recommendation list and recommending the recommendation list to the corresponding enterprise user.
3. The method as claimed in claim 2, wherein the encoding in the process (2) is performed by using n 'bit status registers to encode n' states, each state has its own independent register bit, and only one bit is valid at any time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210439549.5A CN114996566A (en) | 2022-04-25 | 2022-04-25 | Intelligent recommendation system and method for industrial internet platform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210439549.5A CN114996566A (en) | 2022-04-25 | 2022-04-25 | Intelligent recommendation system and method for industrial internet platform |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114996566A true CN114996566A (en) | 2022-09-02 |
Family
ID=83025116
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210439549.5A Pending CN114996566A (en) | 2022-04-25 | 2022-04-25 | Intelligent recommendation system and method for industrial internet platform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114996566A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116383491A (en) * | 2023-03-21 | 2023-07-04 | 北京百度网讯科技有限公司 | Information recommendation method, apparatus, device, storage medium, and program product |
-
2022
- 2022-04-25 CN CN202210439549.5A patent/CN114996566A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116383491A (en) * | 2023-03-21 | 2023-07-04 | 北京百度网讯科技有限公司 | Information recommendation method, apparatus, device, storage medium, and program product |
| CN116383491B (en) * | 2023-03-21 | 2024-05-24 | 北京百度网讯科技有限公司 | Information recommendation method, apparatus, device, storage medium, and program product |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112214685B (en) | Knowledge graph-based personalized recommendation method | |
| CN110083770B (en) | Sequence recommendation method based on deeper feature level self-attention network | |
| CN111737578B (en) | Recommendation method and system | |
| CN111222332A (en) | Commodity recommendation method combining attention network and user emotion | |
| CN112487199B (en) | User characteristic prediction method based on user purchasing behavior | |
| Zhang et al. | Cross-domain recommendation with semantic correlation in tagging systems | |
| CN112016002A (en) | Mixed recommendation method integrating comment text level attention and time factors | |
| CN113343125B (en) | Academic accurate recommendation-oriented heterogeneous scientific research information integration method and system | |
| CN112712418B (en) | Method and device for determining recommended commodity information, storage medium and electronic equipment | |
| CN111563770A (en) | Click rate estimation method based on feature differentiation learning | |
| CN114065048A (en) | Article recommendation method based on multi-different-pattern neural network | |
| CN112256965A (en) | Neural collaborative filtering model recommendation method based on lambdamat | |
| CN112699310A (en) | Cold start cross-domain hybrid recommendation method and system based on deep neural network | |
| Alfarhood et al. | DeepHCF: a deep learning based hybrid collaborative filtering approach for recommendation systems | |
| CN116228368A (en) | Advertisement click rate prediction method based on deep multi-behavior network | |
| Xue et al. | Learning knowledge graph embedding with multi-granularity relational augmentation network | |
| CN114996566A (en) | Intelligent recommendation system and method for industrial internet platform | |
| US20240037133A1 (en) | Method and apparatus for recommending cold start object, computer device, and storage medium | |
| Huang et al. | Broad recommender system: An efficient nonlinear collaborative filtering approach | |
| CN114139066A (en) | Collaborative filtering recommendation system based on graph neural network | |
| CN113344648A (en) | Advertisement recommendation method and system based on machine learning | |
| Zeng et al. | User Personalized Recommendation Algorithm Based on GRU Network Model in Social Networks | |
| Qi et al. | Deep Field‐Aware Interaction Machine for Click‐Through Rate Prediction | |
| Sangeetha et al. | An Enhanced Neural Graph based Collaborative Filtering with Item Knowledge Graph | |
| CN114565436A (en) | Vehicle model recommendation system, method, device and storage medium based on time sequence modeling |
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 |