CN112365302B - Product recommendation network training method, device, equipment and medium - Google Patents
Product recommendation network training method, device, equipment and medium Download PDFInfo
- Publication number
- CN112365302B CN112365302B CN202011467379.9A CN202011467379A CN112365302B CN 112365302 B CN112365302 B CN 112365302B CN 202011467379 A CN202011467379 A CN 202011467379A CN 112365302 B CN112365302 B CN 112365302B
- Authority
- CN
- China
- Prior art keywords
- product
- network
- recommendation
- target
- value
- 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.)
- Active
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
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
-
- 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
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0255—Targeted advertisements based on user history
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Physics & Mathematics (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The disclosure provides a training method, a device, equipment and a medium for a product recommendation network, wherein the method comprises the following steps: inputting user information of a target user, product information of a first target product and product history information of the first target product into a deep recommendation network obtained by pre-training, and outputting a product recommendation value of the first target product corresponding to the target user by the deep recommendation network; inputting the user information of the target user and the product information of the first target product into a product recommendation network to be trained, and outputting a product recommendation predicted value of the first target product corresponding to the target user by the product recommendation network; and adjusting the network parameters of the product recommendation network according to the difference between the product recommendation predicted value and the product recommendation value.
Description
Technical Field
The disclosure relates to the technical field of deep learning, in particular to a training method, a device, equipment and a medium for a product recommendation network.
Background
In different product recommendation scenarios, the cold start problem of new products is often encountered. That is, after a new product is put on the shelf on a platform, due to lack of product history information such as browsing, clicking and the like of a user, the related recommendation technology is difficult to accurately recommend the new product to a corresponding user.
Disclosure of Invention
At least one embodiment of the disclosure provides a training scheme of a product recommendation network, so that a new product is accurately recommended by using the product recommendation network obtained through training.
In a first aspect, a method for training a product recommendation network is provided, the method including:
inputting user information of a target user, product information of a first target product and product history information of the first target product into a deep recommendation network obtained by pre-training, and outputting a product recommendation value of the first target product corresponding to the target user by the deep recommendation network;
inputting the user information of the target user and the product information of the first target product into a product recommendation network to be trained, and outputting a product recommendation predicted value of the first target product corresponding to the target user by the product recommendation network;
and adjusting the network parameters of the product recommendation network according to the difference between the product recommendation predicted value and the product recommendation value.
In a second aspect, a method for recommending products is provided, the method comprising:
determining that a second target product is a new product according to product history information of the second target product;
inputting user information of a target user and product information of the second target product into a product recommendation network, and outputting a product recommendation value of the second target product corresponding to the target user by the product recommendation network; the product recommendation network is obtained by training according to the method provided by the first aspect;
recommending the second target product to the target user based on the product recommendation value.
In a third aspect, there is provided a training apparatus for a product recommendation network, the apparatus comprising:
the product recommendation value output module is used for inputting user information of a target user, product information of a first target product and product history information of the first target product into a deep recommendation network obtained through pre-training, and outputting a product recommendation value of the first target product corresponding to the target user through the deep recommendation network;
the product recommendation prediction value output module is used for inputting the user information of the target user and the product information of the first target product into a product recommendation network to be trained, and outputting the product recommendation prediction value of the first target product corresponding to the target user by the product recommendation network;
and the network parameter adjusting module is used for adjusting the network parameters of the product recommendation network according to the difference between the product recommendation predicted value and the product recommendation value.
In a fourth aspect, there is provided a product recommendation device, the device comprising:
the new product determining module is used for determining that a second target product is a new product according to the product history information of the second target product;
the new product recommendation value output module is used for inputting the user information of the target user and the product information of the second target product into a product recommendation network, and outputting the product recommendation value of the second target product corresponding to the target user by the product recommendation network; the product recommendation network is obtained by training according to the method provided by the first aspect;
and the recommending module is used for recommending the second target product to the target user based on the product recommending value.
In a fifth aspect, a computer device is provided, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the processor implements the method for training a product recommendation network according to any embodiment of the present disclosure, or implements the method for recommending a product according to any embodiment of the present disclosure.
In a sixth aspect, a computer-readable storage medium is provided, on which a computer program is stored, which when executed by a processor implements the method for training a product recommendation network according to any of the embodiments of the present disclosure, or implements the method for recommending a product according to any of the embodiments of the present disclosure.
According to the technical scheme, in at least one embodiment of the disclosure, the user information of the target user, the product information of the first target product and the product history information are input into a pre-trained deep recommendation network, so that the product recommendation value of the target user corresponding to the first target product is obtained and is used as the label value; and inputting the user information of the target user and the product information of the first target product into a product recommendation network to be trained to obtain a product recommendation predicted value, and adjusting network parameters of the product recommendation network based on the difference between the product recommendation predicted value and the label value. In the training mode, based on a pre-trained deep recommendation network, a label value of a training product recommendation network is obtained according to product history information of a target product, and network parameters of the product recommendation network are adjusted based on the label value to train the product recommendation network. The product recommendation network obtained by training in the method can be matched and recommended based on the input user information and the product information, so that the defect that the new product lacks product historical information is overcome, and the new product lacking the product historical information is recommended more accurately.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and together with the description, serve to explain the principles of the disclosure.
FIG. 1 is a flow diagram illustrating a method for training a product recommendation network in accordance with an exemplary embodiment;
FIG. 2 is a flow diagram illustrating a method for deriving product recommendation predictions in accordance with an exemplary embodiment;
FIG. 3 is a flow diagram illustrating a method of recommending products in accordance with an exemplary embodiment;
FIG. 4 is a schematic diagram of a training apparatus for a product recommendation network, according to an example embodiment;
FIG. 5 is a schematic diagram of a training apparatus for yet another product recommendation network, shown in accordance with an exemplary embodiment;
FIG. 6 is a schematic diagram illustrating a product recommendation device according to an exemplary embodiment.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The specific manner described in the following exemplary embodiments does not represent all aspects consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.
The terminology used in the present disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. As used in this disclosure and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.
It is to be understood that although the terms first, second, third, etc. may be used herein to describe various information, such information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present disclosure. The word "if" as used herein may be interpreted as "at … …" or "when … …" or "in response to a determination", depending on the context.
In each product recommendation scenario, a cold start problem for new products is often encountered. For example, after a new product is put on shelf for the first time on a platform, due to lack of product history information such as user browsing, clicking or conversion information, the existing recommendation algorithm is difficult to accurately match the new product with a user, so that the new product is accurately recommended to the corresponding user. Illustratively, the recommended products may be articles, commodities, financial products, and the like.
Based on the above, the present disclosure provides a training method for a product recommendation network. The method comprises the steps of obtaining a label value required by training a product recommendation network according to product history information based on a pre-trained deep recommendation network, and adjusting network parameters of the product recommendation network based on the difference between the label value and an output value of the product recommendation network. The product recommendation network obtained by training can realize more accurate recommendation values for new products and users on the basis of not inputting product historical information, thereby realizing accurate recommendation of the new products.
In order to make the training method of the product recommendation network provided by the present disclosure clearer, the following describes in detail the implementation process of the solution provided by the present disclosure with reference to the accompanying drawings and specific embodiments.
Referring to fig. 1, fig. 1 is a flowchart illustrating a training method of a product recommendation network according to an embodiment provided in the present disclosure. As shown in fig. 1, the process includes:
The user information is information related to the user himself. Illustratively, the information can be the age, height, weight, sex, shopping preference, shopping habit and the like of the user. The product information is inherent information of the product. Illustratively, the classification information, labeling information, etc. that a product determines when it is on the shelf on the platform, and these information do not change over the time of the shelf. The product history information is information related to the user generated after the product is put on shelf. Illustratively, the product history information may be user browsing information, user click information, purchase record information, click amount information, etc. after the product is placed on shelf.
In the implementation, before training the product recommendation network, the deep recommendation network needs to be obtained by training in advance. The deep recommendation network is a deep learning network capable of learning. In some optional embodiments, the Deep recommendation network may include a Deep neural network DNN, Wide & Deep model, or ESMM model. The deep recommendation network obtained through training can determine the product recommendation value of the user corresponding to the product according to the user information of the user, the product information of the product and the product history information of the product. Wherein the product recommendation value is used to characterize the accuracy of recommending the corresponding product to the customer. For example, a higher product recommendation value indicates a higher user demand for the product.
For example, user information, product information and product history information can be collected as training samples based on relevant data of old products on the platform; and the information of whether the user clicks the corresponding product or not can be used as a label value in the training sample. For example, if the user clicks on a product, the tag value is 1, and if the user does not click on a product, the tag value is 0. And based on the obtained training sample, taking the user information, the product information and the product history information as the input of the deep recommendation network to be trained, and adjusting the network parameters of the deep recommendation network based on a predetermined label value to finally obtain the trained deep recommendation network. The training process of the above deep recommendation network is a routine practice in the art and is not described in detail.
After the deep recommendation network is obtained through pre-training, the user information of the target user, the product information of the first target product and the product history information of the first target product can be input into the deep recommendation network, and the product recommendation value of the target user corresponding to the first target product is output through the deep recommendation network. Wherein the target user is a user who needs to make a product recommendation, the first target product is a product available for recommendation, and the first target product has corresponding product history information (e.g., the first target product is an old product).
The deep recommendation network can more accurately obtain the product recommendation value of the first target product corresponding to the target user due to the fact that three input information, namely user information, product information and product history information, are completely input. The product recommendation value obtained by the deep recommendation network can be used as a basis for recommending the first target product to the target user. For example, based on the product recommendation value of the target user corresponding to the first target product, it may be determined that the first target product is set at a specific position of the recommended product. For example, it may be determined to place a first target product in the first place of the recommended product based on the product recommendation value.
In this embodiment, the product recommendation value obtained by the deep recommendation network is used as a tag value for training the product recommendation network, so as to implement training of the product recommendation network based on the tag value.
And 102, inputting the user information of the target user and the product information of the first target product into a product recommendation network to be trained, and outputting a product recommendation predicted value of the first target product corresponding to the target user by the product recommendation network.
In the step, the user information of the target user and the product information of the first target product can be used as input information of a product recommendation network to be trained, and the product recommendation network outputs the product recommendation predicted value of the target user corresponding to the first target product according to the input two information. The product recommendation network is a deep learning network capable of training and learning. In some possible implementations, the product recommendation network may include a Deep neural network DNN, Wide & Deep model, or ESMM model.
Because the input of the product recommendation network only comprises the user information and the product information but lacks the product history information, the product recommendation predicted value obtained by the product recommendation network according to the two input information is not accurate enough, and the product recommendation predicted value is different from the product recommendation value obtained by the deep recommendation network. This step may adjust network parameters of the product recommendation network based on a difference between the two to train the product recommendation network.
In the embodiment of the disclosure, based on a pre-trained deep recommendation network, a product recommendation value of a target user corresponding to a first target product is obtained as a label value of a training product recommendation network according to three information, namely input user information, product information and product history information; and adjusting network parameters of the product recommendation network according to the difference between the product recommendation predicted value and the label value obtained by the product recommendation network according to the user information and the product information, so as to realize training and learning of the product recommendation network. The product recommendation network obtained by the training mode can obtain the corresponding product recommendation value only according to the user information and the product information, namely the product recommendation network can obtain the product recommendation value of the user corresponding to the new product under the condition of lacking product history information, so that the accurate recommendation of the new product can be realized.
In some optional embodiments, the product recommendation network comprises: a fitting sub-network and at least one base sub-network. The fitting sub-network may comprise a linear regression or a regression tree, among others. The base subnetwork may comprise: a Deep neural network DNN, Wide & Deep model, or ESMM model.
In the case where the product recommendation network includes a fitting sub-network and at least one base sub-network, as shown in FIG. 2, the specific implementation of step 102 may include the steps of:
In the case that the product recommendation network includes only one basic sub-network, this step may input the user information of the target user and the product information of the first target product as complete input information into the basic sub-network and obtain the corresponding intermediate recommendation value.
In the case that the product recommendation network includes at least two basic sub-networks, the step may input the user information of the target user and the product information of the first target product as complete input information into each of the at least two basic sub-networks, respectively, and obtain corresponding at least two intermediate recommendation values from each of the basic sub-networks.
Under the condition that at least one intermediate recommended value is obtained, the obtained intermediate recommended value can be input into the fitting sub-network, and the product recommended predicted value of the target user corresponding to the first target product is obtained by the fitting sub-network according to the input intermediate recommended value.
In the case where the product recommendation network includes a fitting sub-network and at least one base sub-network, network parameters of the fitting sub-network and the base sub-network included in the product recommendation network may be adjusted according to a difference between the product recommendation prediction value and the product recommendation value when adjusting network parameters of the product recommendation network.
In some optional embodiments, at least one of the base sub-networks included in the product recommendation network is a base network trained in advance. The training process of the basic sub-network is similar to that of the deep recommendation network, and the only difference is that the basic sub-network only inputs user information and product information in the training process. Because the basic sub-network is the intermediate recommended value obtained according to the two kinds of information, namely the user information and the product information, the intermediate recommended value obtained by the basic sub-network is not accurate enough compared with the product recommended value obtained by the deep recommended network according to the three kinds of information, namely the product information and the product history information, namely the difference exists. Therefore, in this embodiment, the intermediate recommended values obtained by the pre-trained basic sub-network can be further used as the input of the fitting sub-network, and the fitting sub-network can obtain the final product recommended predicted value according to the input intermediate recommended values.
In the above embodiment, a preliminary intermediate recommended value is obtained by a pre-trained basic sub-network according to the user information and the product information, and the intermediate recommended value is further processed by a fitting sub-network, so as to reduce a difference between the intermediate recommended value and a product recommended value obtained by a deep recommendation network. Therefore, the product recommendation network obtained by training in the method can obtain a recommendation value between the product and the user which is closer to the recommendation value obtained by the deep recommendation network, and a new product can be recommended better.
In one possible implementation, the product recommendation network includes at least two basic sub-networks, and different basic sub-networks are pre-trained based on different training samples. Wherein the different training samples are two training samples that are not identical. For example, the training sample 1 may include user information 1 and product information 1; wherein, the user information 1 may include randomly determined 8 types of user information; the product information 1 includes therein 6 types of product information determined at random. The training sample 2 can include user information 2 and product information 2; the user information 2 comprises 8 types of randomly determined user information; the product information 2 includes therein 6 types of product information determined at random. Training sample 1 and training sample 2 may be referred to as two different training samples.
In the implementation manner, different basic sub-networks can be trained based on different training samples, and the basic sub-networks trained by the method can learn more features from different training samples. Therefore, the product recommendation network can obtain different intermediate recommendation values based on different basic sub-networks, and the fitting sub-network can more accurately obtain the product recommendation prediction value according to the plurality of intermediate recommendation values.
In some alternative embodiments, in case the product recommendation network comprises at least two basic sub-networks, the network structure of the different basic sub-networks is different. For example, the product recommendation network includes three basic sub-networks, namely a basic sub-network 1, a basic sub-network 2 and a basic sub-network 3, wherein the basic sub-network 1 can be a deep neural network DNN network; the base subnetwork 2 can be Wide & Deep model; the base subnetwork 3 may be an ESMM model.
In the above embodiment, the network structures of different basic sub-networks are different, so that the different basic sub-networks can extract richer features of the input information to obtain different intermediate recommendation values. Furthermore, the fitting sub-network can further process different intermediate recommended values, and finally a more accurate product recommended predicted value is obtained.
Only the training process of the product recommendation network is described above, and after the product recommendation network meeting the requirements is obtained through training, the product recommendation network can be used for recommending the product. In the following embodiments, a product recommendation method is disclosed, as shown in fig. 3, the method includes the following steps:
In this embodiment, the second target product is a product to be recommended. This step may determine whether the second target product is a new product according to product history information of the second target product. For example, a second target product may be determined to be a new product in the event that the product history information does not exist for the second target product; in the case where the product history information exists for the second target product, it is determined that the second target product is an old product.
In some alternative embodiments, the new product may be determined based on the amount of product history information for the second target product. For example, in the case that the amount of the product history information of the second target product is less than the preset threshold, the second target product may be determined as a new product. For example, the threshold may be preset as: 20 browsing records; if the browsing records of the second target product are 10, determining that the second target product is a new product; and if the browsing records of the second target product are 30, determining that the second target product is an old product. Therefore, products with less product historical information can be classified into new products, and corresponding product recommendation values can be obtained by utilizing the product recommendation network obtained through training.
In some optional embodiments, in the case that the second target product is determined to be an old product, the product recommendation value of the second target product may be obtained directly by using the deep recommendation network.
And under the condition that the product recommendation value of the target user corresponding to the second target product is obtained, recommending the second target product to the target user according to the product recommendation value.
In some optional embodiments, the product recommendation values of different new products corresponding to the target user may be obtained based on the product recommendation network; and obtaining product recommendation values of different old products corresponding to the target user based on the deep recommendation network. Further, the product recommendation values of the new products and the product recommendation values of the old products are sorted according to the recommendation values, and the corresponding new products or the corresponding old products are recommended to the target users according to the sorting. The product recommendation mode can realize the recommendation of old products and the reasonable recommendation of new products.
As shown in fig. 4, the present disclosure provides a training apparatus for a product recommendation network, which may perform the training method for a product recommendation network according to any embodiment of the present disclosure. The apparatus may include a product recommendation value output module 401, a product recommendation prediction value output module 402, and a network parameter adjustment module 403. Wherein:
a product recommendation value output module 401, configured to input user information of a target user, product information of a first target product, and product history information of the first target product into a deep recommendation network obtained through pre-training, and output a product recommendation value of the first target product corresponding to the target user by the deep recommendation network;
a product recommendation prediction value output module 402, configured to input the user information of the target user and the product information of the first target product into a product recommendation network to be trained, and output, by the product recommendation network, a product recommendation prediction value of the first target product corresponding to the target user;
a network parameter adjusting module 403, configured to adjust a network parameter of the product recommendation network according to a difference between the product recommendation prediction value and the product recommendation value.
Optionally, the product recommendation network comprises: fitting a sub-network and at least one base sub-network;
the product recommendation prediction value output module 402, as shown in fig. 5, includes:
the intermediate recommended value output sub-module 501 is configured to input the user information of the target user and the product information of the first target product into the at least one basic sub-network, and output at least one intermediate recommended value by the at least one basic sub-network;
and a product recommendation prediction value output sub-module 502, configured to input the at least one intermediate recommendation value into the fitting sub-network, and output, by the fitting sub-network, a product recommendation prediction value of the first target product corresponding to the target user.
Optionally, the network parameter adjusting module 403, when configured to adjust the network parameter of the product recommendation network according to a difference between the product recommendation prediction value and the product recommendation value, includes:
adjusting network parameters of the fitting sub-network and at least one base sub-network according to a difference between the product recommendation prediction value and the product recommendation value.
Optionally, the at least one base subnetwork is pre-trained; the network parameter adjusting module 403, when configured to adjust the network parameter of the product recommendation network according to the difference between the product recommendation prediction value and the product recommendation value, includes:
and adjusting the network parameters of the fitting sub-network according to the difference between the product recommendation prediction value and the product recommendation value.
Optionally, in a case that the number of the at least one base sub-network is at least two, the at least two base sub-networks are pre-trained based on different training samples.
Optionally, in a case that the number of the at least one basic sub-network is at least two, the at least one basic sub-network includes basic sub-networks of at least two different network structures.
As shown in fig. 6, the present disclosure provides a product recommendation device that can perform the product recommendation method of any embodiment of the present disclosure. The apparatus may include a new product determination module 601, a new product recommendation value output module 602, and a recommendation module 603. Wherein:
a new product determining module 601, configured to determine that a second target product is a new product according to product history information of the second target product;
a new product recommendation value output module 602, configured to input user information of a target user and product information of the second target product into a product recommendation network, and output, by the product recommendation network, a product recommendation value of the second target product corresponding to the target user; the product recommendation network is obtained by training according to a training method in any embodiment of the specification;
a recommending module 603, configured to recommend the second target product to the target user based on the product recommendation value.
Optionally, the new product determining module 601, when configured to determine that a second target product is a new product according to product history information of the second target product, includes:
determining that the second target product is a new product if the quantity of the product history information of the second target product is less than a preset threshold.
For the device embodiments, since they substantially correspond to the method embodiments, reference may be made to the partial description of the method embodiments for relevant points. The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of at least one embodiment of the present disclosure. One of ordinary skill in the art can understand and implement it without inventive effort.
The present disclosure also provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor is capable of implementing the training method of the product recommendation network according to any embodiment of the present disclosure, or implementing the product recommendation method according to any embodiment of the present disclosure when executing the program.
The present disclosure also provides a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, is capable of implementing the method for training a product recommendation network of any embodiment of the present disclosure, or implementing the method for product recommendation of any embodiment of the present disclosure.
The non-transitory computer readable storage medium may be, among others, ROM, Random Access Memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, and the like, and the present disclosure is not limited thereto.
Other embodiments of the present disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
The above description is only exemplary of the present disclosure and is not intended to limit the present disclosure, so that any modification, equivalent replacement, or improvement made within the spirit and principle of the present disclosure should be included in the scope of the present disclosure.
Claims (18)
1. A method of training a product recommendation network, the method comprising:
inputting user information of a target user, product information of a first target product and product history information of the first target product into a deep recommendation network obtained by pre-training, and outputting a product recommendation value of the first target product corresponding to the target user by the deep recommendation network;
inputting the user information of the target user and the product information of the first target product into a product recommendation network to be trained, and outputting a product recommendation predicted value of the first target product corresponding to the target user by the product recommendation network;
adjusting network parameters of the product recommendation network according to the difference between the product recommendation prediction value and the product recommendation value;
the product recommendation network includes: fitting a sub-network and at least one base sub-network;
the inputting the user information of the target user and the product information of the first target product into a product recommendation network to be trained, and outputting the product recommendation predicted value of the first target product corresponding to the target user by the product recommendation network includes:
inputting the user information of the target user and the product information of the first target product into the at least one basic sub-network, and outputting at least one intermediate recommendation value by the at least one basic sub-network;
and inputting the at least one intermediate recommendation value into the fitting sub-network, and outputting a product recommendation prediction value of the first target product corresponding to the target user by the fitting sub-network.
2. The method of claim 1, the adjusting a network parameter of the product recommendation network according to a difference between the product recommendation prediction value and the product recommendation value, comprising:
adjusting network parameters of the fitting sub-network and at least one base sub-network according to a difference between the product recommendation prediction value and the product recommendation value.
3. The method of claim 1, the at least one base subnetwork is pre-trained;
the adjusting the network parameters of the product recommendation network according to the difference between the product recommendation prediction value and the product recommendation value comprises:
and adjusting the network parameters of the fitting sub-network according to the difference between the product recommendation prediction value and the product recommendation value.
4. The method of claim 3, wherein at least two of the base sub-networks are pre-trained based on different training samples in the case that the number of the at least one base sub-network is at least two.
5. The method according to any of claims 1-4, comprising in the at least one base sub-network at least two base sub-networks of different network structures, in case the number of the at least one base sub-network is at least two.
6. The method of any of claims 1-4, the base subnetwork comprising: a Deep neural network DNN, Wide & Deep model, or ESMM model.
7. The method of any of claims 1-4, the fitting a sub-network comprising: linear regression or regression trees.
8. A method of product recommendation, the method comprising:
determining that a second target product is a new product according to product history information of the second target product;
inputting user information of a target user and product information of the second target product into a product recommendation network, and outputting a product recommendation value of the second target product corresponding to the target user by the product recommendation network; the product recommendation network is trained according to the method of any one of claims 1-7;
recommending the second target product to the target user based on the product recommendation value.
9. The method of claim 8, the determining that a second target product is a new product according to product history information for the second target product, comprising:
determining that the second target product is a new product if the quantity of the product history information of the second target product is less than a preset threshold.
10. An apparatus for training a product recommendation network, the apparatus comprising:
the product recommendation value output module is used for inputting user information of a target user, product information of a first target product and product history information of the first target product into a deep recommendation network obtained through pre-training, and outputting a product recommendation value of the first target product corresponding to the target user through the deep recommendation network;
the product recommendation prediction value output module is used for inputting the user information of the target user and the product information of the first target product into a product recommendation network to be trained, and outputting the product recommendation prediction value of the first target product corresponding to the target user by the product recommendation network;
the network parameter adjusting module is used for adjusting the network parameters of the product recommendation network according to the difference between the product recommendation predicted value and the product recommendation value;
the product recommendation network includes: fitting a sub-network and at least one base sub-network;
the product recommendation prediction value output module comprises:
the intermediate recommendation value output sub-module is used for inputting the user information of the target user and the product information of the first target product into the at least one basic sub-network and outputting at least one intermediate recommendation value by the at least one basic sub-network;
and the product recommendation prediction value output sub-module is used for inputting the at least one intermediate recommendation value into the fitting sub-network, and outputting the product recommendation prediction value of the first target product corresponding to the target user by the fitting sub-network.
11. The apparatus of claim 10, the network parameter adjustment module, when configured to adjust the network parameters of the product recommendation network based on a difference between the product recommendation prediction value and the product recommendation value, comprises:
adjusting network parameters of the fitting sub-network and at least one base sub-network according to a difference between the product recommendation prediction value and the product recommendation value.
12. The apparatus of claim 10, the at least one base subnetwork is pre-trained; the network parameter adjusting module, when configured to adjust the network parameter of the product recommendation network according to a difference between the product recommendation prediction value and the product recommendation value, includes:
and adjusting the network parameters of the fitting sub-network according to the difference between the product recommendation prediction value and the product recommendation value.
13. The apparatus of claim 12, wherein at least two of the base subnetworks are pre-trained based on different training samples if the number of the at least one base subnetwork is at least two.
14. The apparatus of any of claims 10-13, in a case that the number of the at least one base sub-network is at least two, including base sub-networks of at least two different network structures in the at least one base sub-network.
15. A product recommendation device, the device comprising:
the new product determining module is used for determining that a second target product is a new product according to the product history information of the second target product;
the new product recommendation value output module is used for inputting the user information of the target user and the product information of the second target product into a product recommendation network, and outputting the product recommendation value of the second target product corresponding to the target user by the product recommendation network; the product recommendation network is trained according to the method of any one of claims 1-7;
and the recommending module is used for recommending the second target product to the target user based on the product recommending value.
16. The device of claim 15, the new product determination module, when configured to determine that a second target product is a new product based on product history information for the second target product, comprises:
determining that the second target product is a new product if the quantity of the product history information of the second target product is less than a preset threshold.
17. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1-7 or implements the method of any of claims 8-9 when executing the program.
18. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the method of any one of claims 1 to 7, or carries out the method of any one of claims 8 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011467379.9A CN112365302B (en) | 2020-12-14 | 2020-12-14 | Product recommendation network training method, device, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011467379.9A CN112365302B (en) | 2020-12-14 | 2020-12-14 | Product recommendation network training method, device, equipment and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112365302A CN112365302A (en) | 2021-02-12 |
| CN112365302B true CN112365302B (en) | 2021-03-23 |
Family
ID=74534511
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011467379.9A Active CN112365302B (en) | 2020-12-14 | 2020-12-14 | Product recommendation network training method, device, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112365302B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112989213B (en) * | 2021-05-19 | 2022-05-20 | 腾讯科技（深圳）有限公司 | Content recommendation method, device and system, electronic equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104239374A (en) * | 2013-06-21 | 2014-12-24 | 伊姆西公司 | Method and device for file recommendation |
| CN111695965A (en) * | 2020-04-26 | 2020-09-22 | 清华大学 | Product screening method, system and equipment based on graph neural network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11205210B2 (en) * | 2018-12-20 | 2021-12-21 | Leafly Holdings, Inc. | System for selection of regulated products |
-
2020
- 2020-12-14 CN CN202011467379.9A patent/CN112365302B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104239374A (en) * | 2013-06-21 | 2014-12-24 | 伊姆西公司 | Method and device for file recommendation |
| CN111695965A (en) * | 2020-04-26 | 2020-09-22 | 清华大学 | Product screening method, system and equipment based on graph neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112365302A (en) | 2021-02-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110046952B (en) | Recommendation model training method and device, and recommendation method and device | |
| CN108960719B (en) | Method and device for selecting products and computer readable storage medium | |
| CN108205766A (en) | Information-pushing method, apparatus and system | |
| CN110288484B (en) | Insurance classification user recommendation method and system based on big data platform | |
| EP3324352A1 (en) | Testing system | |
| CN110069699B (en) | Ranking model training method and device | |
| CN110096617B (en) | Video classification method and device, electronic equipment and computer-readable storage medium | |
| CN109784654A (en) | Task creating method, device, computer equipment and storage medium | |
| CN108022116B (en) | Method, system and terminal equipment for modeling user | |
| CN113946754A (en) | User portrait based rights and interests recommendation method, device, equipment and storage medium | |
| CN108053351A (en) | Intelligent college entrance will commending system and recommendation method | |
| CN108596765A (en) | A kind of Electronic Finance resource recommendation method and device | |
| CN110825974B (en) | Recommendation system content ordering method and device | |
| CN111861679A (en) | Commodity recommendation method based on artificial intelligence | |
| CN112365302B (en) | Product recommendation network training method, device, equipment and medium | |
| CN112053205A (en) | Product recommendation method and device through robot emotion recognition | |
| US20130346334A1 (en) | Determining course equivalence | |
| CN108595580B (en) | News recommendation method, device, server and storage medium | |
| CN111125561A (en) | Network heat display method and device | |
| KR101640867B1 (en) | A method and a system for providing user-customized learning course based on machine learning | |
| CN115511582B (en) | Commodity recommendation system and method based on artificial intelligence | |
| US20230015090A1 (en) | Systems and Methods for Dynamically Classifying Products and Assessing Applicability of Product Regulations | |
| CN114840659A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN112734142A (en) | Resource learning path planning method and device based on deep learning | |
| CN113590673A (en) | Data heat degree statistical method based on block chain deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |