CN112150238A - Deep neural network-based commodity recommendation method and system - Google Patents
Deep neural network-based commodity recommendation method and system Download PDFInfo
- Publication number
- CN112150238A CN112150238A CN202010930144.2A CN202010930144A CN112150238A CN 112150238 A CN112150238 A CN 112150238A CN 202010930144 A CN202010930144 A CN 202010930144A CN 112150238 A CN112150238 A CN 112150238A
- Authority
- CN
- China
- Prior art keywords
- user
- commodity
- layer
- neural network
- recommendation
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention belongs to the technical field of intelligent recommendation, and discloses a commodity recommendation method and system based on a deep neural network, wherein the commodity recommendation method based on the deep neural network comprises the following steps: firstly, an input layer acquires characteristics of a user and a commodity; the embedded layer performs primary processing on the characteristics of the user and the commodity; secondly, the user-commodity interaction layer carries out user characteristic and commodity characteristic interaction; potential feature depth extraction is carried out on the residual error network layer; and finally, the output layer predicts the commodity recommendation required by the user through a sigmoid activation function. The method can effectively reduce the non-related commodity recommendation of the platform to the user, and utilizes the deep residual error network to replace a common neural network in the neural collaborative filtering recommendation algorithm, thereby capturing the high-order nonlinear features in the user-article relation data, solving the problem that the extracted high-order nonlinear features are insufficient due to the simpler neural network used in the current recommendation algorithm, and achieving better recommendation effect.
Description
Technical Field
The invention belongs to the technical field of intelligent recommendation, and particularly relates to a commodity recommendation method and system based on a deep neural network.
Background
At present, with the continuous expansion of the electronic commerce scale, the number and the types of the commodities are rapidly increased, and customers need to spend a great deal of time to find the commodities which the customers want to buy. This process of browsing through large amounts of unrelated information and products will undoubtedly result in a constant loss of consumers who are overwhelmed by the problem of information overload. Many platforms have their own recommendation systems, which can analyze and recommend their own platform-related commodities according to user's usual behavior habits and browsing records, but also record the user's usual accidental operations, and merchant platforms often recommend some commodities that are not of interest to themselves or commodities that are being browsed at one time, because the recommendation algorithm cannot deeply capture the high-order nonlinear interaction relationship in the user-item interaction data, the platform recommends the user's irrelevant commodities.
Two neural network-based collaborative filtering frameworks on the market at present, one is a neural collaborative filtering framework (NCF), the model combines matrix decomposition (MF) and multi-layer perceptron (MLP) to learn the user-item interaction characteristics, and the linear modeling advantage of the MF and the nonlinear modeling advantage of the MLP are unified; the other is a neural collaborative filtering recommendation algorithm (ONCF) based on outer product, which represents the user-item relationship in the form of outer product and learns the user-item interaction characteristics through a convolutional neural network.
Although the NCF and the ONCF adopt a relatively simple neural network to learn the user-item interaction characteristics, the complex and high-sparsity characteristic diagram is not enough to learn the complex and high-order user-item nonlinear characteristics deeply, and the existing problems of the recommendation algorithm in the market at present cannot be solved.
Through the above analysis, the problems and defects of the prior art are as follows: the existing simple neural network has the problems of gradient disappearance and gradient explosion, so that the problem of insufficient extraction of high-order nonlinear features is caused, the recommendation effect is poor, and the recommendation of non-related commodities is more.
The difficulty in solving the above problems and defects is: although the traditional corresponding solution is initialization and regularization of data, the problem of gradient is solved, the depth is increased, but another problem is brought, namely the degradation problem of network performance is solved, the depth is increased, and the error rate is increased.
The significance of solving the problems and the defects is as follows: the residual error network is designed to solve the degradation problem, and simultaneously, the gradient problem is also solved, so that the performance of the network is improved. Therefore, the deep residual error network can solve the problem that the MLP in the neural matrix model is difficult to capture highly complex nonlinear potential features in the user-item interaction data.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a commodity recommendation method and system based on a deep neural network.
The invention is realized in such a way, a commodity recommendation method based on a deep neural network comprises the following steps:
firstly, an input layer acquires characteristics of a user and a commodity; the embedded layer performs primary processing on the characteristics of the user and the commodity; secondly, the user-commodity interaction layer carries out user characteristic and commodity characteristic interaction; potential feature depth extraction is carried out on the residual error network layer; and finally, the output layer predicts the commodity recommendation required by the user through a sigmoid activation function.
Further, the deep neural network-based commodity recommendation method comprises the following steps:
acquiring specification data of users and commodities, adopting one-hot coding on the characteristics of the users and the products on an input layer, representing the users and the commodities as hidden characteristic vectors, and discretizing information characteristics;
step two, mapping the one-hot encoded binary sparse vector of the input layer into a dense implicit vector through an embedding layer;
step three, obtaining a user-article interactive relation graph with rich semanteme by carrying out outer product on the user implicit vector and the article vector of the embedded layer;
step four, adopting a ResNet50 depth residual error network to perform potential feature depth extraction;
and fifthly, predicting the commodity required by the user by utilizing the output layer.
Further, in step three, the outer product formula is as follows:
wherein p isuIs a set of users, q, for which user u has a past behavior on item iiIs a collection of items of interest for which user u may be present in item i.
Further, in step four, the ResNet50 depth residual network is composed of five parts, conv1, conv2_ x, conv3_ x, conv4_ x and conv5_ x, wherein the number of layers of each part is 2, 3 × 3, 3 × 4, 3 × 6 and 3 × 3; each part is composed of a shortcut structure map, and each layer is a convolution layer.
Further, in the fifth step, the predicting the commodity required by the user by using the output layer includes:
connecting the output of the full connection layer and the output of the residual error network, wherein the number of output nodes of the full connection layer is 1, and predicting by using a sigmoid function as a final activation function; continuously learning the model by the minimum objective function, the formula is as follows:
wherein the content of the first and second substances,a predicted value representing the possibility that the user u is related to the item i, namely the user's liking degree to the item; y represents a true value, 1 represents that the user u is related to the item i, and 0 represents that the user u is not related to the item i; the above-mentionedAnd Y can be optimized by training using a random gradient descent.
The invention also aims to provide a merchant platform recommendation terminal for implementing the deep neural network-based commodity recommendation method.
Another object of the present invention is to provide a deep neural network-based commodity recommendation system for implementing the deep neural network-based commodity recommendation method, the deep neural network-based commodity recommendation system including:
the input layer is used for acquiring characteristics of the user and the commodity;
the embedded layer is used for carrying out primary processing on the characteristics of the user and the commodity;
the user-commodity interaction layer is used for carrying out user characteristic and commodity characteristic interaction;
the residual error network layer is used for carrying out potential feature depth extraction;
and the output layer is used for predicting the commodity recommendation required by the user through the sigmoid activation function.
7. A computer device, characterized in that the computer device comprises a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to carry out the steps of:
the input layer acquires the characteristics of the user and the commodity; the embedded layer performs primary processing on the characteristics of the user and the commodity;
the user-commodity interaction layer carries out user characteristic and commodity characteristic interaction; potential feature depth extraction is carried out on the residual error network layer;
and the output layer predicts the commodity recommendation required by the user through a sigmoid activation function.
8. A computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of:
the input layer acquires the characteristics of the user and the commodity; the embedded layer performs primary processing on the characteristics of the user and the commodity;
the user-commodity interaction layer carries out user characteristic and commodity characteristic interaction; potential feature depth extraction is carried out on the residual error network layer;
and the output layer predicts the commodity recommendation required by the user through a sigmoid activation function.
By combining all the technical schemes, the invention has the advantages and positive effects that: the method can deeply capture the high-order nonlinear features in the user-article relation data, and solves the problem that the extraction of the high-order nonlinear features is insufficient due to the fact that a neural network used in the conventional recommendation method is simple, so that a good recommendation effect is achieved, and the recommendation of some non-related commodities is avoided.
The method can effectively reduce the non-related commodity recommendation of the platform to the user, and utilizes the deep residual error network to replace a common neural network in the neural collaborative filtering recommendation algorithm, thereby capturing the high-order nonlinear features in the user-article relation data, solving the problem that the extracted high-order nonlinear features are insufficient due to the simpler neural network used in the current recommendation algorithm, and achieving better recommendation effect.
Compared with the conventional neural collaborative filtering recommendation algorithm, the commodity recommendation algorithm based on the deep neural network solves the degradation problem, solves the gradient problem at the same time, and further improves the performance of the network. Therefore, the deep residual error network solves the problem that the MLP in the neural matrix model is difficult to capture highly complex nonlinear potential features in user-article interaction data, can effectively reduce non-related commodity recommendation of the platform to the user, and greatly improves the performance of a commodity recommendation algorithm.
The technical effect or experimental effect of comparison comprises the following steps:
the experimental effect is as follows:
(1) description of experimental data set. The invention uses the data set disclosed in the field of recommendation systems to verify the effectiveness of the method proposed herein, which is also the more MovieLens1M used in movie data sets. This data set is also widely used for verifying the movie scoring data set of the performance effect of the CF algorithm, and there are 100 ten thousand interaction records. Relevant data statistics for this data set are shown in table 1:
table 1: relevant data statistics of a data set
Data set | Number of users | Number of movies | Score of | Degree of sparseness |
MovieLens-1M | 6040 | 3900 | 1000209 | 95.80% |
(2) And selecting an evaluation scheme and an index. The invention adopts leave one method to evaluate, namely: for each user, the last interaction is used as the test set, and the rest of the interactions are used as the training set. Since it takes too much time to rank all items for each user in the evaluation process, following a general strategy, 100 items are randomly drawn without interacting with the user, and the performance of the ranking list is measured by Hit Rate (HR) and Normalized Discount Cumulative Gain (NDCG). HR is to see how many of the first ten items in the item recommendation list are from the test set, and NDCG is to see how relevant the recommendation list is. The two indexes of each test user are calculated, the average score is obtained, and the larger the values of the two indexes are, the better the recommendation performance of the model is.
(3) Introduction of the reference method. In order to verify the effectiveness of the depth residual error network collaborative filtering model (DRNCF) model proposed herein, the present invention selects some reference methods for comparison tests.
User-CF: the method is characterized in that the similarity between users is calculated by analyzing the behavior records of the users, and then the users with high similarity with the target user and the objects which do not interact with the target user are recommended to the target user, so that the method is a more social recommendation algorithm.
Item-CF: the method is an algorithm with personalized recommendation, which is used for recommending the target user the items with high similarity to the items which are liked by the target user before by analyzing the behavior records of the user and calculating the similarity between the items.
GMF: the method maps the user and the article to the same shared potential feature space, and the interaction relationship is represented by the inner product of the corresponding potential feature vectors.
MLP: this approach replaces the simple inner product with MLP to model the interaction between the user and the item. The MLP structure of the experimental part is a tower-shaped structure containing three hidden layers.
NeuMF: is the most advanced algorithm for item recommendation, and combines hidden layers of GMF and MLP to learn the interaction relationship between the user and the item.
(4) Experimental parameters: the data set is a large number of movie files, MovieLens1M is used in the movie data set, the number K of recommendation lists is 10, the size of an embedded layer is 224, the learning rate is 0.001, and a small batch Adam optimization model is used.
The trend of DRNCF performance with other baseline methods for fifty training sessions before the Movielens-1M data set with the same embedded layer size and recommendation list length is shown in FIG. 4.
As can be seen from the above table, the DRNCF model outperforms other baseline methods in HR on the published data set.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments of the present application will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained from the drawings without creative efforts.
Fig. 1 is a schematic diagram of a deep neural network-based commodity recommendation method according to an embodiment of the present invention.
Fig. 2 is a flowchart of a deep neural network-based commodity recommendation method according to an embodiment of the present invention.
FIG. 3 is a schematic structural diagram of a deep neural network-based commodity recommendation system according to an embodiment of the present invention;
in the figure: 1. an input layer; 2. an embedding layer; 3. a user-commodity interaction layer; 4. a residual network layer; 5. and (5) outputting the layer.
FIG. 4 is a graph of the trend of the DRNCF and baseline method of the invention in the HR and NDCG on Movielens-1M. In fig. 4: (a) HR @ 10; (b) NDCG @ 10.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
In order to solve the problems in the prior art, the invention provides a deep neural network-based commodity recommendation method, which is described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the deep neural network-based commodity recommendation method provided in the embodiment of the present invention includes:
firstly, an input layer acquires characteristics of a user and a commodity; the embedded layer performs primary processing on the characteristics of the user and the commodity; secondly, the user-commodity interaction layer carries out user characteristic and commodity characteristic interaction; potential feature depth extraction is carried out on the residual error network layer; and finally, the output layer predicts the commodity recommendation required by the user through a sigmoid activation function.
In the residual error network, the numerical value represents the predicted value of the satisfaction degree of the commodity required by the output user, such as: the User-Item interaction diagram is an interaction relation diagram of a User requirement characteristic and a screen characteristic; 1-D Conv1_ x, Conv2_ x, …, Conv5_ x denote the number of levels of the residual network,a predicted value representing a degree of satisfaction of a user with a certain commodity.
As shown in fig. 2, the deep neural network-based commodity recommendation method provided in the embodiment of the present invention includes the following steps:
s101, acquiring specification data of users and commodities, adopting one-hot coding on the characteristics of the users and the commodities on an input layer, representing the users and the commodities as hidden characteristic vectors, and discretizing information characteristics;
s102, mapping the one-hot coded binary sparse vector of the input layer into a dense implicit vector through an embedding layer;
s103, obtaining a user-article interactive relation graph with rich semanteme by carrying out outer product on the user implicit vector and the article vector of the embedded layer;
s104, potential feature depth extraction is carried out by adopting a ResNet50 depth residual error network;
and S105, predicting the commodity required by the user by utilizing the output layer.
In step S103, the outer product formula provided in the embodiment of the present invention is as follows:
wherein p isuIs a set of users, q, for which user u has a past behavior on item iiIs a collection of items of interest for which user u may be present in item i.
In step S104, the ResNet50 depth residual error network provided in the embodiment of the present invention is composed of five parts, i.e., conv1, conv2_ x, conv3_ x, conv4_ x and conv5_ x, where the number of layers in each part is 2, 3 × 3, 3 × 4, 3 × 6, and 3 × 3; each part is composed of a shortcut structure map, and each layer is a convolution layer.
In step S105, the predicting a commodity required by a user by using an output layer according to an embodiment of the present invention includes:
connecting the output of the full connection layer and the output of the residual error network, wherein the number of output nodes of the full connection layer is 1, and predicting by using a sigmoid function as a final activation function; continuously learning the model by the minimum objective function, the formula is as follows:
wherein the content of the first and second substances,a predicted value representing the possibility that the user u is related to the item i, namely the user's liking degree to the item; y represents a true value, 1 represents that the user u is related to the item i, and 0 represents that the user u is not related to the item i; the above-mentionedAnd Y can be optimized by training using a random gradient descent.
As shown in fig. 3, the deep neural network-based commodity recommendation system according to the embodiment of the present invention includes:
the input layer 1 is used for acquiring characteristics of users and commodities;
the embedded layer 2 is used for carrying out primary processing on the characteristics of the user and the commodity;
the user-commodity interaction layer 3 is used for carrying out user characteristic and commodity characteristic interaction;
a residual error network layer 4 for performing potential feature depth extraction;
and the output layer 5 is used for predicting the commodity recommendation required by the user through the sigmoid activation function.
The technical solution of the present invention is further illustrated by the following specific examples.
Example (b):
a commodity recommendation algorithm based on a deep neural network is divided into five steps as shown in figure 1, wherein the first step is an input layer and a user and commodity feature acquisition stage; the second step is an embedding layer, a user and commodity feature initial processing stage; the third step is a user-commodity interaction layer, and a user characteristic and commodity characteristic interaction stage; a fourth step is a residual error network layer, a potential feature stage is extracted deeply, and a deep residual error network is adopted to solve the problem that highly complex nonlinear potential features in user-commodity interaction data are difficult to capture in an MLP (multi-level label) in a neural matrix model; and in the stage of predicting the commodity required by the user, the fifth step is an output layer, and the commodity recommendation required by the user is predicted by taking the sigmoid function as an activation function.
The method comprises the following specific steps:
step 1: input layer, user and commodity feature obtaining stage.
Firstly, acquiring specification data of users and commodities, then adopting one-hot coding on the characteristics of the users and the commodities on an input layer, and enabling the users and the commodities to be represented by implicit characteristic vectors, so that each user or commodity is represented by only using continuous numbers of 0 and 1, information characteristics are clearly discretized, and the subsequent extraction of user-commodity interaction relationship is facilitated to maintain.
Step 2: embedding layer, initial processing stage of user and commodity characteristics.
The one-hot encoded binarization sparse vector of the input layer is mapped into a dense implicit vector through the embedding layer, so that the sparsity is reduced, and the calculation amount is reduced.
And step 3: user-commodity interaction layer, user characteristic and commodity characteristic interaction phase.
And carrying out outer product operation on the user implicit vector and the item vector of the embedded layer to obtain a user-item interactive relation graph with rich semanteme. This is different from the MF model and the neural matrix model, which take into account the correlation between different embedding dimensions, and which encode more information than the inner product operation in the MF model and the cascade operation in the neural matrix. More importantly, the two-dimensional matrix format is the same as the image, so that the neural network can learn the features conveniently. The formula is as follows:
and 4, step 4: residual network layer, deep extracting potential characteristic stage.
The method adopts ResNet50 in a depth residual network, wherein the network consists of five parts including conv1, conv2_ x, conv3_ x, conv4_ x and conv5_ x, and the number of layers of each part is 2, 3 × 3, 3 × 4, 3 × 6 and 3 × 3. Each part is composed of a shortcut structure diagram, and each layer is not a full connection layer any more but a convolution layer. The ResNet50 replaces an MLP network, the problem of insufficient extraction of high-order nonlinear implicit features of the user-article is solved, under the condition of the same input and the same network depth, the ResNet50 uses a lot of parameters which are less than that of the MLP, and the overall performance of the model is greatly improved.
And 5: output layer, stage of predicting goods required by user
The value of Y is taken as a label-1 to indicate that user u is associated with item i, otherwise it is 0. Predicted scoreIt represents the likelihood that user u is associated with item i. The fully-connected layer is connected with the output of the residual error network, the number of output nodes of the fully-connected layer is 1, and the sigmoid function is used as the final activation function for prediction due to the fact that the prediction is carried out through two classes (0/1). The model is then continuously learned through a minimum objective function, which is formulated as follows:
whereinRepresents a predicted value of the user's liking of the item, and Y represents a true value. And training optimization can be performed by using random gradient descent (SGD).
The trend of DRNCF performance with other baseline methods for fifty training sessions before the Movielens-1M data set with the same embedded layer size and recommendation list length is shown in FIG. 4.
In the description of the present invention, "a plurality" means two or more unless otherwise specified; the terms "upper", "lower", "left", "right", "inner", "outer", "front", "rear", "head", "tail", and the like, indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, are only for convenience in describing and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed in a particular orientation, and be operated, and thus, should not be construed as limiting the invention. Furthermore, the terms "first," "second," "third," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
The above description is only for the purpose of illustrating the present invention and the appended claims are not to be construed as limiting the scope of the invention, which is intended to cover all modifications, equivalents and improvements that are within the spirit and scope of the invention as defined by the appended claims.
Claims (9)
1. A deep neural network-based commodity recommendation method is characterized by comprising the following steps:
the input layer acquires the characteristics of the user and the commodity;
the embedded layer performs primary processing on the characteristics of the user and the commodity;
the user-commodity interaction layer carries out user characteristic and commodity characteristic interaction; potential feature depth extraction is carried out on the residual error network layer;
and the output layer predicts the commodity recommendation required by the user through a sigmoid activation function.
2. The deep neural network-based commodity recommendation method according to claim 1, wherein the deep neural network-based commodity recommendation method comprises the steps of:
acquiring specification data of users and commodities, adopting one-hot coding on the characteristics of the users and the products on an input layer, representing the users and the commodities as hidden characteristic vectors, and discretizing information characteristics;
step two, mapping the one-hot encoded binary sparse vector of the input layer into a dense implicit vector through an embedding layer;
step three, obtaining a user-article interactive relation graph with rich semanteme by carrying out outer product on the user implicit vector and the article vector of the embedded layer;
step four, adopting a ResNet50 depth residual error network to perform potential feature depth extraction;
and fifthly, predicting the commodity required by the user by utilizing the output layer.
3. The deep neural network-based commodity recommendation method according to claim 2, wherein in step three, the outer product formula is as follows:
wherein p isuIs a set of users, q, for which user u has a past behavior on item iiIs a collection of items of interest for which user u may be present in item i.
4. The deep neural network-based commodity recommendation method according to claim 2, wherein in step four, the ResNet50 deep residual network is composed of five parts, i.e., conv1, conv2_ x, conv3_ x, conv4_ x and conv5_ x, and the number of layers of each part is 2, 3, 4, 3, 6 and 3; each part is composed of a shortcut structure map, and each layer is a convolution layer.
5. The deep neural network-based commodity recommendation method according to claim 2, wherein in the fifth step, the predicting the commodity required by the user by using the output layer comprises:
connecting the output of the full connection layer and the output of the residual error network, wherein the number of output nodes of the full connection layer is 1, and predicting by using a sigmoid function as a final activation function; continuously learning the model by the minimum objective function, the formula is as follows:
wherein the content of the first and second substances,a predicted value representing the possibility that the user u is related to the item i, namely the user's liking degree to the item; y represents a true value, 1 represents that the user u is related to the item i, and 0 represents that the user u is not related to the item i; the above-mentionedAnd Y can be optimized by training using a random gradient descent.
6. A deep neural network-based commodity recommendation system for implementing the deep neural network-based commodity recommendation method according to claim 1, wherein the deep neural network-based commodity recommendation system comprises:
the input layer is used for acquiring characteristics of the user and the commodity;
the embedded layer is used for carrying out primary processing on the characteristics of the user and the commodity;
the user-commodity interaction layer is used for carrying out user characteristic and commodity characteristic interaction;
the residual error network layer is used for carrying out potential feature depth extraction;
and the output layer is used for predicting the commodity recommendation required by the user through the sigmoid activation function.
7. A merchant platform recommendation terminal implementing the deep neural network-based commodity recommendation method of claim 1.
8. A computer device, characterized in that the computer device comprises a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to carry out the steps of:
the input layer acquires the characteristics of the user and the commodity; the embedded layer performs primary processing on the characteristics of the user and the commodity;
the user-commodity interaction layer carries out user characteristic and commodity characteristic interaction; potential feature depth extraction is carried out on the residual error network layer;
and the output layer predicts the commodity recommendation required by the user through a sigmoid activation function.
9. A computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of:
the input layer acquires the characteristics of the user and the commodity; the embedded layer performs primary processing on the characteristics of the user and the commodity;
the user-commodity interaction layer carries out user characteristic and commodity characteristic interaction; potential feature depth extraction is carried out on the residual error network layer;
and the output layer predicts the commodity recommendation required by the user through a sigmoid activation function.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010930144.2A CN112150238A (en) | 2020-09-07 | 2020-09-07 | Deep neural network-based commodity recommendation method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010930144.2A CN112150238A (en) | 2020-09-07 | 2020-09-07 | Deep neural network-based commodity recommendation method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112150238A true CN112150238A (en) | 2020-12-29 |
Family
ID=73889342
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010930144.2A Pending CN112150238A (en) | 2020-09-07 | 2020-09-07 | Deep neural network-based commodity recommendation method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112150238A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113283671A (en) * | 2021-06-22 | 2021-08-20 | 润联软件系统(深圳)有限公司 | Method and device for predicting replenishment quantity, computer equipment and storage medium |
CN113407819A (en) * | 2021-05-20 | 2021-09-17 | 桂林电子科技大学 | Sequence recommendation method, system and storage medium based on residual error network |
CN113627598A (en) * | 2021-08-16 | 2021-11-09 | 重庆大学 | Twin self-encoder neural network algorithm and system for accelerated recommendation |
CN113761378A (en) * | 2021-09-14 | 2021-12-07 | 上海任意门科技有限公司 | Content ordering method, computing device and computer-readable storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107491540A (en) * | 2017-08-24 | 2017-12-19 | 济南浚达信息技术有限公司 | A kind of combination depth Bayesian model and the film of collaboration Heterogeneous Information insertion recommend method |
CN108458716A (en) * | 2018-02-02 | 2018-08-28 | 北京交通大学 | A kind of electric vehicle charging air navigation aid based on the prediction of charging pile dynamic occupancy |
CN108536856A (en) * | 2018-04-17 | 2018-09-14 | 重庆邮电大学 | Mixing collaborative filtering film recommended models based on two aside network structure |
CN110120227A (en) * | 2019-04-26 | 2019-08-13 | 天津大学 | A kind of depth stacks the speech separating method of residual error network |
-
2020
- 2020-09-07 CN CN202010930144.2A patent/CN112150238A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107491540A (en) * | 2017-08-24 | 2017-12-19 | 济南浚达信息技术有限公司 | A kind of combination depth Bayesian model and the film of collaboration Heterogeneous Information insertion recommend method |
CN108458716A (en) * | 2018-02-02 | 2018-08-28 | 北京交通大学 | A kind of electric vehicle charging air navigation aid based on the prediction of charging pile dynamic occupancy |
CN108536856A (en) * | 2018-04-17 | 2018-09-14 | 重庆邮电大学 | Mixing collaborative filtering film recommended models based on two aside network structure |
CN110120227A (en) * | 2019-04-26 | 2019-08-13 | 天津大学 | A kind of depth stacks the speech separating method of residual error network |
Non-Patent Citations (1)
Title |
---|
吴敏: "深度学习框架下充电桩的个性化推荐算法研究", 《中国优秀硕士学位论文全文数据库工程科技II辑》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113407819A (en) * | 2021-05-20 | 2021-09-17 | 桂林电子科技大学 | Sequence recommendation method, system and storage medium based on residual error network |
CN113407819B (en) * | 2021-05-20 | 2022-06-17 | 桂林电子科技大学 | Sequence recommendation method, system and storage medium based on residual error network |
CN113283671A (en) * | 2021-06-22 | 2021-08-20 | 润联软件系统(深圳)有限公司 | Method and device for predicting replenishment quantity, computer equipment and storage medium |
CN113627598A (en) * | 2021-08-16 | 2021-11-09 | 重庆大学 | Twin self-encoder neural network algorithm and system for accelerated recommendation |
CN113627598B (en) * | 2021-08-16 | 2022-06-07 | 重庆大学 | Twin self-encoder neural network algorithm and system for accelerating recommendation |
CN113761378A (en) * | 2021-09-14 | 2021-12-07 | 上海任意门科技有限公司 | Content ordering method, computing device and computer-readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110717098B (en) | Meta-path-based context-aware user modeling method and sequence recommendation method | |
CN111797321B (en) | Personalized knowledge recommendation method and system for different scenes | |
CN111460130B (en) | Information recommendation method, device, equipment and readable storage medium | |
CN112150238A (en) | Deep neural network-based commodity recommendation method and system | |
CN111461841B (en) | Article recommendation method, device, server and storage medium | |
CN111626832B (en) | Product recommendation method and device and computer equipment | |
CN106599226A (en) | Content recommendation method and content recommendation system | |
CN110008397B (en) | Recommendation model training method and device | |
CN112487199B (en) | User characteristic prediction method based on user purchasing behavior | |
CN108805598A (en) | Similarity information determines method, server and computer readable storage medium | |
CN113761359B (en) | Data packet recommendation method, device, electronic equipment and storage medium | |
CN111695965A (en) | Product screening method, system and equipment based on graph neural network | |
CN111523421A (en) | Multi-user behavior detection method and system based on deep learning and fusion of various interaction information | |
CN114443958A (en) | Recommendation method, recommendation system and recommendation system training method | |
CN115438787A (en) | Training method and device of behavior prediction system | |
Ji et al. | Regularized singular value decomposition in news recommendation system | |
CN115358809A (en) | Multi-intention recommendation method and device based on graph comparison learning | |
CN110348906A (en) | A kind of improvement Method of Commodity Recommendation based on polymorphic type implicit feedback | |
US20240037133A1 (en) | Method and apparatus for recommending cold start object, computer device, and storage medium | |
CN112541806A (en) | Recommendation method and device based on heterogeneous information network | |
CN112036987A (en) | Method and device for determining recommended commodities | |
CN114417161B (en) | Virtual article time sequence recommendation method, device, medium and equipment based on special-purpose map | |
CN115391555A (en) | User-perceived knowledge map recommendation system and method | |
CN116071119A (en) | Model-agnostic inverse fact interpretation method based on multi-behavior recommendation model | |
CN114861072A (en) | Graph convolution network recommendation method and device based on interlayer combination mechanism |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201229 |