CN112417293A - Information pushing method and system, model training method and related equipment - Google Patents
Information pushing method and system, model training method and related equipment Download PDFInfo
- Publication number
- CN112417293A CN112417293A CN202011397968.4A CN202011397968A CN112417293A CN 112417293 A CN112417293 A CN 112417293A CN 202011397968 A CN202011397968 A CN 202011397968A CN 112417293 A CN112417293 A CN 112417293A
- Authority
- CN
- China
- Prior art keywords
- model
- target domain
- domain prediction
- training
- user
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000012549 training Methods 0.000 title claims abstract description 133
- 238000000034 method Methods 0.000 title claims abstract description 75
- 238000012545 processing Methods 0.000 claims abstract description 17
- 238000012795 verification Methods 0.000 claims description 30
- 230000006870 function Effects 0.000 claims description 18
- 238000003860 storage Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 11
- 238000002360 preparation method Methods 0.000 claims description 8
- 238000010586 diagram Methods 0.000 description 26
- 230000008569 process Effects 0.000 description 14
- 230000008901 benefit Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 239000002453 shampoo Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- General Engineering & Computer Science (AREA)
- Finance (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an information pushing method and system, a model training method and related equipment, and relates to the field of data processing. The information pushing method comprises the following steps: acquiring a target domain prediction model which is sent by an offline system and completes training, wherein the target domain prediction model is obtained by training by using a prediction result of a source domain prediction model and user training data of a target domain, and the source domain prediction model is obtained by training by using the user training data of the source domain; and predicting the data to be tested of the user in the target domain by using the target domain prediction model so as to obtain an information pushing result of the corresponding user. Therefore, when a new service or a new application scene is on line, the embodiment of the invention can quickly and accurately provide a corresponding prediction model.
Description
Technical Field
The present invention relates to the field of data processing, and in particular, to an information pushing method and system, a model training method, and related devices.
Background
Within an enterprise, there is often a series of product matrices, where each product has user data for a particular business scenario. In the personalized recommendation method in the related art, model training is usually performed based on actual self service scene data.
Disclosure of Invention
In a part of service scenes, some products in a growth period are often not enough in user data, so that when model training in the service scenes is carried out, the models are difficult to obtain good generalization capability, and accordingly the personalized effect is poor.
The embodiment of the invention aims to solve the technical problem that: how to improve the accuracy of user recommendation in a new service scene.
According to a first aspect of some embodiments of the present invention, there is provided an information pushing method, including: acquiring a target domain prediction model which is sent by an offline system and completes training, wherein the target domain prediction model is obtained by training by using a prediction result of a source domain prediction model and user training data of a target domain, and the source domain prediction model is obtained by training by using the user training data of the source domain; and predicting the data to be tested of the user in the target domain by using the target domain prediction model so as to obtain an information pushing result of the corresponding user.
In some embodiments, the target domain prediction model which is sent by the offline system and has been trained and the coding value of the target domain prediction model are obtained periodically, and the information pushing method further includes: under the condition that the code value of the obtained prediction network model is different from the code value of the currently used target domain prediction model, the version of the obtained target domain prediction model is verified; and under the condition that the online verification of the obtained target domain prediction model is passed, replacing the currently used target domain prediction model with the obtained target domain prediction model, wherein the online verification comprises version verification.
In some embodiments, the online verification further comprises model verification, and the information pushing method further comprises: and under the condition that the value of the preset parameter of the obtained target domain prediction model is within a preset range, performing model verification on the obtained target domain prediction model.
In some embodiments, the obtained target domain prediction model is a cured graph file in which parameters of the target domain prediction model determined by training are converted into constants.
In some embodiments, the data to be tested of the user includes characteristics of the user and characteristics of the product, and the information pushing result of the corresponding user is a result of whether the product is recommended for the user.
According to a second aspect of some embodiments of the present invention, there is provided a model training method for information push, including: training a source domain prediction model by using user training data of a source domain; respectively inputting source domain characteristic data and target domain characteristic data corresponding to the same user into a source domain prediction model and a target domain prediction model to obtain a source domain prediction result and a target domain prediction result corresponding to the user; and adjusting parameters of the target domain prediction model according to the difference between the source domain prediction result and the target domain prediction result and the difference between the marking value corresponding to the user and the target domain prediction result.
In some embodiments, adjusting the parameters of the target domain prediction model comprises: and adjusting parameters of the target domain prediction model based on a loss function of the target domain prediction model, wherein the loss function of the target domain prediction model comprises the cross entropy of the source domain prediction result and the target domain prediction result and the cross entropy of the mark value corresponding to the user and the target domain prediction result.
In some embodiments, the complexity of the target domain prediction model is higher than the complexity of the source domain prediction model.
In some embodiments, the number of dimensions of the input data to the target domain prediction model is greater than the number of dimensions of the input data to the source domain prediction model.
In some embodiments, the model training method further comprises: acquiring first characteristic data and second characteristic data corresponding to each of a plurality of users, wherein the second characteristic data corresponding to the same user comprises characteristics of partial dimensions of the first characteristic data; training a first preparatory model by using the first characteristic data; inputting first characteristic data and second characteristic data corresponding to the same user into a first preliminary model and a second preliminary model respectively to obtain a first preliminary model prediction result and a second preliminary model prediction result corresponding to the user; adjusting parameters of the second preparation model according to the difference between the prediction result of the first preparation model and the prediction result of the second preparation model and the difference between the marking value corresponding to the user and the prediction result of the first preparation model; and taking the adjusted second preparation model as a source domain prediction model.
In some embodiments, the first preparatory model and the second preparatory model have the same network model structure except that the input layers are different.
In some embodiments, the source domain prediction result and the target domain prediction result corresponding to the user are recommendation results associated with the same item.
According to a third aspect of some embodiments of the present invention, there is provided an information pushing apparatus, including: the system comprises an acquisition module, a storage module and a processing module, wherein the acquisition module is configured to acquire a target domain prediction model which is sent by an offline system and completes training, the target domain prediction model is acquired by utilizing a prediction result of a source domain prediction model and user training data of a target domain for training, and the source domain prediction model is acquired by utilizing user training data of the source domain for training; the prediction module is configured to predict the data to be tested of the user in the target domain by using the target domain prediction model so as to obtain an information pushing result of the corresponding user.
According to a fourth aspect of some embodiments of the present invention, there is provided a model training apparatus for information push, including: a source domain training module configured to train a source domain prediction model using user training data of a source domain; the target domain training module is configured to input the source domain characteristic data and the target domain characteristic data corresponding to the same user into the source domain prediction model and the target domain prediction model respectively to obtain a source domain prediction result and a target domain prediction result corresponding to the user; and adjusting parameters of the target domain prediction model according to the difference between the source domain prediction result and the target domain prediction result and the difference between the marking value corresponding to the user and the target domain prediction result.
According to a fifth aspect of some embodiments of the present invention, there is provided an information push system including: an information push device; and the model training device is used for pushing information.
According to a sixth aspect of some embodiments of the present invention, there is provided an information pushing apparatus, including: a memory; and a processor coupled to the memory, the processor configured to execute any one of the foregoing information pushing methods based on instructions stored in the memory.
According to a seventh aspect of some embodiments of the present invention, there is provided a model training apparatus for information push, including: a memory; and a processor coupled to the memory, the processor configured to execute any one of the aforementioned model training methods for information push based on instructions stored in the memory.
According to an eighth aspect of some embodiments of the present invention, there is provided a computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements any of the aforementioned information push methods, or any of the aforementioned model training methods for information push.
Some embodiments of the above invention have the following advantages or benefits. The embodiment of the invention effectively utilizes cross-domain data information, aggregates the knowledge that a plurality of data islands can learn, and assists the training of the target domain prediction model by combining the training result of the source domain prediction model, thereby further improving the generalization capability of the model. Therefore, when a new service or a new application scene is on line, the embodiment of the invention can quickly and accurately provide a corresponding prediction model.
Other features of the present invention and advantages thereof will become apparent from the following detailed description of exemplary embodiments thereof, which proceeds with reference to the accompanying drawings.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 illustrates a flow diagram of an information push method according to some embodiments of the inventions.
Fig. 2 illustrates a flow diagram of a model training method for information push according to some embodiments of the present invention.
FIG. 3 illustrates a flow diagram of a pre-training method according to some embodiments of the invention.
FIG. 4 illustrates a flow diagram of a model verification method according to some embodiments of the inventions.
Fig. 5 shows a schematic structural diagram of an information pushing device according to some embodiments of the invention.
Fig. 6 shows a schematic structural diagram of a model training apparatus for information push according to some embodiments of the present invention.
Fig. 7 illustrates a schematic structural diagram of an information push system according to some embodiments of the invention.
FIG. 8 shows a block diagram of a data processing apparatus according to some embodiments of the present invention.
FIG. 9 shows a block diagram of a data processing apparatus according to further embodiments of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The relative arrangement of the components and steps, the numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.
Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description.
Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate.
In all examples shown and discussed herein, any particular value should be construed as merely illustrative, and not limiting. Thus, other examples of the exemplary embodiments may have different values.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.
FIG. 1 illustrates a flow diagram of an information push method according to some embodiments of the inventions. As shown in fig. 1, the information push method of this embodiment includes steps S102 to S104.
In step S102, a trained prediction network model transmitted by the offline system is obtained, where the prediction network model is obtained by training using the prediction result of the teacher network model of the prediction network model and the user training data of the target domain, and the teacher network model of the prediction model is obtained by training using the user training data of the source domain.
In some embodiments, a "domain" represents a different business scenario. The source domain is, for example, a mature service scenario with a large amount of available user data, and the target domain is, for example, a new service scenario with a small amount of user data.
For example, a business hosted by a company is e-commerce, and the e-commerce platform has sufficient user purchase data. If the likelihood of a user purchasing a good is predicted, thereby recommending the good for the user, the prediction model may be trained using the user purchase data. Then, the company expands the financial services, and part of users of the original e-commerce platform open the financial services. However, since users who set up such services are limited, it is difficult to obtain a model with high accuracy if the prediction model is trained only depending on the financial service data of the users. At this time, the user purchase data may be regarded as data of the source domain, and the user financial transaction data may be regarded as data of the target domain. Training of the target domain prediction model is aided by a training result with data of the source domain.
In the related art, a commonly used training method of a neural network model is as follows: and training the difference between the prediction result of the training data and the marking value according to the model. However, since the target domain has a problem of insufficient data amount, a source domain prediction model obtained by further training with training data having sufficient data amount in the training process is considered. When the target domain prediction model is trained, the prediction accuracy of input data of the target domain prediction model is considered, and the difference between the prediction result of the target domain prediction model and the prediction result of the source domain prediction model is also considered, so that the target domain prediction model can learn the learning knowledge of the source domain prediction model. Therefore, even when the data amount of the target domain is small, the target domain prediction model with high accuracy can be obtained. The specific training method for the target domain prediction model will be further described in the following embodiments.
In some embodiments, the obtained target domain prediction model is a cured graph file in which parameters of the target domain prediction model determined by training are converted into constants. Therefore, the online system obtains a lighter model file, and the online efficiency of a new version model is improved.
In step S104, the data to be tested of the user in the target domain is predicted by using the prediction network model, so as to obtain an information pushing result for the corresponding user.
In some embodiments, the user data to be tested includes characteristics of the user, characteristics of the product, environmental characteristics (e.g., time characteristics, characteristics of other users, characteristics of related products, characteristics of platform activities, etc.), and the like.
And the information pushing result of the corresponding user is the result of whether the product is recommended for the user. For example, if the data to be tested of the user a includes the characteristics of the user a and the characteristics of a certain shampoo, the information pushing result indicates whether the shampoo is recommended for the user a. In some embodiments, the prediction network model outputs a judgment probability, and a recommendation result is determined according to a comparison result between the judgment probability and a preset probability.
After the recommendation result is determined, the recommendation result can be sent to the terminal of the user in a preset information format by using a push interface between the front-end application module and the background server.
The embodiment effectively utilizes cross-domain data information, aggregates the knowledge that a plurality of data islands can learn, and assists the training of the target domain prediction model by combining the training result of the source domain prediction model, thereby further improving the generalization capability of the model. Therefore, when a new service or a new application scene is on line, the embodiment of the invention can quickly and accurately provide a corresponding prediction model.
An embodiment of the model training method for information push of the present invention is described below with reference to fig. 2.
Fig. 2 illustrates a flow diagram of a model training method for information push according to some embodiments of the present invention. As shown in fig. 2, the model training method for information push of this embodiment includes steps S202 to S206.
In some embodiments, prior to performing the training process, a user's exposure log, click log, product content forward index, user portrait feature log, and the like are collected. After collecting these data, a plurality of pieces of user data are obtained by performing data fusion, for example, by identification such as a device number, and removing dirty data lacking valid features. In some embodiments, positive samples and negative samples can be extracted from the obtained data according to a preset proportion; in addition, samples that do not meet preset conditions, such as samples that are often below a certain threshold for browsing of the product, may be filtered out.
In step S202, a source domain prediction model is trained using user training data for the source domain.
In step S204, the source domain characteristic data and the target domain characteristic data corresponding to the same user are respectively input into the source domain prediction model and the target domain prediction model, so as to obtain a source domain prediction result and a target domain prediction result corresponding to the user. Namely, inputting source domain characteristic data corresponding to a certain user into a source domain prediction model to obtain a source domain prediction result; and inputting the target domain characteristic data corresponding to the same user into a target domain prediction model to obtain a target domain prediction result.
In step S206, parameters of the target domain prediction model are adjusted according to a difference between the source domain prediction result and the target domain prediction result and a difference between the labeled value corresponding to the user and the target domain prediction result.
In some embodiments, parameters of the target domain prediction model are adjusted based on a loss function of the target domain prediction model, wherein the loss function of the target domain prediction model includes a cross entropy of the source domain prediction result and the target domain prediction result, and a cross entropy of the user corresponding token value and the target domain prediction result. For example, equation (1) is used as a loss function of the target domain prediction model.
L＝CE(y,pred)+αCE(q,pred) (1)
In formula (1), L represents the value of the loss function; CE (, indicates that cross entropy is calculated for two variables within parentheses; y represents a flag value; pred represents the target domain prediction result; q represents a source domain prediction result; α represents a preset parameter.
In some embodiments, the value of q is determined by the softmax layer represented by equation (2).
In the formula (2), qiRepresenting the probability corresponding to the ith class in the classification result of the source domain prediction model; z is a radical ofiRepresenting the result which corresponds to the ith type and is input to the softmax layer; j denotes the identity of the respective class given by the source domain prediction model, zjRepresenting the result which corresponds to the jth class and is input to the softmax layer; t represents a preset "temperature value" parameter for the tableShowing the softening degree of the prediction result of the source domain prediction model. In some embodiments, T-10, where the inventors tested, this value resulted in better training. In determining q corresponding to each classiThen, the maximum value is defined as the value of q.
In some embodiments, the source domain prediction result and the target domain prediction result corresponding to the user are recommendation results associated with the same item.
For example, a user has both user purchase data on an e-commerce platform as a source domain and financial transaction data on a financial transaction platform as a target domain. And setting recommendation for a certain mobile phone. The source domain prediction result corresponding to the user is, for example, whether to recommend a mobile phone for the user, and the target domain prediction result corresponding to the user is, for example, whether to recommend a financial service for the user to purchase the mobile phone by stages. If the source domain prediction result and the target domain prediction result are both recommended for the same user and the same item, the output values of the source domain prediction model and the target domain prediction model may be considered to be the same.
In some embodiments, the complexity of the target domain prediction model is higher than the complexity of the source domain prediction model. The complexity of the model is determined, for example, by the number of layers of the model, the number of layers with a predetermined mechanism (e.g., attention mechanism), the number of parameters, and so on.
In some embodiments, the number of dimensions of the input data to the target domain prediction model is greater than the number of dimensions of the input data to the source domain prediction model.
In the related art, the complexity of a model trained later is often lower than that of a model trained earlier, or the model trained later processes data of a lower dimension than that of the model trained earlier. This is to facilitate the post-trained model to meet the requirement of lightweight operation, for example, when the server-side model is transplanted into a mobile terminal for operation, the computing power, the storage power and the data processing power of the mobile terminal need to be considered. However, the embodiment of the present invention is applied to the scenario of data cross-domain learning, so the complexity of the target domain prediction model may be higher than that of the source domain prediction model, and the input data of the target domain prediction model may also be more complex data. Therefore, even for a new service scene, a more complex information pushing thought can be realized.
In some embodiments, to further improve the efficiency of training the target domain prediction model, the source domain prediction model uses as few input dimensions as possible. In order to improve the training efficiency and ensure the training accuracy, in some embodiments, the source domain prediction model is obtained through a pre-training process. An embodiment of the pre-training method of the present invention is described below with reference to fig. 3. The training idea of the embodiment is similar to the idea of training the target domain prediction model, and the training accuracy of one model is improved by means of the training result of the other model.
FIG. 3 illustrates a flow diagram of a pre-training method according to some embodiments of the invention. As shown in fig. 3, the pre-training method of this embodiment includes steps S302 to S310.
In step S302, first feature data and second feature data corresponding to each of a plurality of users are obtained, where the second feature data corresponding to the same user includes features of partial dimensions of the first feature data.
In some embodiments, the first characteristic data and the second characteristic data are both data of the source domain, differing in the number of characteristics of the two. For example, the first feature data is 1000-dimensional user purchase data, and the second feature data takes partial dimensions thereof to form 100-dimensional user purchase data.
In step S304, a first preliminary model is trained using the first feature data.
In step S306, the first feature data and the second feature data corresponding to the same user are input into the first preliminary model and the second preliminary model, respectively, and a first preliminary model prediction result and a second preliminary model prediction result corresponding to the user are obtained.
In step S308, parameters of the second preliminary model are adjusted according to a difference between the first preliminary model prediction result and the second preliminary model prediction result and a difference between the user-associated tag value and the first preliminary model prediction result.
For example, with reference to the training process of the target domain prediction model, the gap can be embodied by cross entropy. In some embodiments, the parameters of the second preparatory model are adjusted based on a loss function of the second preparatory model, wherein the loss function of the second preparatory model includes a cross entropy of the first preparatory model prediction result and the second preparatory model prediction result, and a cross entropy of the user's corresponding token value and the second preparatory model prediction result. For example, equation (3) is used as a loss function of the target domain prediction model.
Lpre＝(1-λ)CE(y′,pred′)+λCE(q′,pred′) (3)
In the formula (3), LpreA value representing a loss function; CE (, indicates that cross entropy is calculated for two variables within parentheses; y' represents a marker value; pred' represents the second preliminary model prediction result; q' represents the first preliminary model prediction result.
The second preliminary network, although having fewer input dimensions, also learns the training results of the first preliminary network obtained by the training of data of more dimensions during the training process. Therefore, the second preliminary network is also enabled to have higher prediction accuracy.
In some embodiments, the first preparatory model and the second preparatory model have the same network model structure except that the input layers are different. Thus, the training process of the second preparatory model may be made more focused on knowledge extraction of unused features.
In step S310, the adjusted second preliminary model is used as a source domain prediction model. After a number of adjustment iterations, the second preparatory model completes training.
In some embodiments, the second preliminary model that has completed training may also be tested. And if the test accuracy is greater than the preset value, taking the second preparation model as a source domain prediction model. If the test accuracy is not greater than the preset value, retraining can be selected; or the input features of the second preliminary model may be considered to be insufficient to characterize the user, requiring re-selection of features as input features of the second preliminary model.
After the second preliminary model is obtained through the pre-training process and the source domain prediction model is further determined, the source domain prediction model can use fewer input features and has the prediction accuracy rate equivalent to that of a model with multi-dimensional feature representation, and therefore the training efficiency of the target domain prediction model is indirectly improved.
In some embodiments, the training process and the information pushing process of the target domain prediction model may be deployed in an offline system and an online system, respectively. The off-line system can periodically update the trained target domain prediction model through the data accumulated in the service process and send the model to the on-line system for application.
In some embodiments, the online system may also look ahead for the target domain prediction model before updating it. An embodiment of the model validation method of the present invention is described below with reference to FIG. 4.
FIG. 4 illustrates a flow diagram of a model verification method according to some embodiments of the inventions. As shown in fig. 4, the model verification method of this embodiment includes steps S402 to S408.
In step S402, the trained target domain prediction model and the encoding value of the target domain prediction model sent by the offline system are periodically obtained. In some embodiments, the encoded value is an MD5 encoded value.
In step S404, in the case where the obtained encoding value of the prediction network model is different from the encoding value of the currently used target domain prediction model, the version verification of the obtained target domain prediction model is passed.
For example, an offline system sends a version of the target domain prediction model that was most recently trained twice. After the first transmission, the online system has used it online. When sending for the second time, if the online system repeatedly executes the same model again, the system efficiency is imaged and system resources are wasted. Therefore, by verifying the code value, the situation of repeated online is avoided, and system resources are saved.
In step S406, an online verification result for the acquired target domain prediction model is determined based on the version verification result.
In some embodiments, the online verification further includes model verification, which is used to verify whether the value of the obtained preset parameter of the target domain prediction model is within a preset range, for example, to see whether the key parameter is empty, etc. Therefore, errors of the sending object or transmission errors can be found in time, and the stability of the system is improved.
In step S408, in the case where the online verification of the acquired target domain prediction model is passed, the acquired target domain prediction model is used instead of the currently used target domain prediction model.
Through the verification process, the stability of the model online process can be improved.
An embodiment of the information pushing apparatus is described below with reference to fig. 5.
Fig. 5 shows a schematic structural diagram of an information pushing device according to some embodiments of the invention. As shown in fig. 5, the information pushing apparatus 500 of this embodiment includes: the obtaining module 5100 is configured to obtain a target domain prediction model which is sent by an offline system and is trained by using a prediction result of a source domain prediction model and user training data of a target domain, and the target domain prediction model is obtained by training by using the user training data of the source domain; the prediction module 5200 is configured to predict, by using the target domain prediction model, data to be tested of the user in the target domain to obtain an information pushing result for the corresponding user.
In some embodiments, the obtaining module 5100 is further configured to periodically obtain the trained target domain prediction model and the encoded value of the target domain prediction model sent by the offline system; the information pushing device 500 further comprises: the verification module 5300 is configured to verify the version of the acquired target domain prediction model if the encoding value of the acquired prediction network model is different from the encoding value of the currently used target domain prediction model; and under the condition that online verification of the obtained target domain prediction model is passed, replacing the currently used target domain prediction model with the obtained target domain prediction model, wherein the online verification comprises the version verification.
In some embodiments, the online verification further includes model verification, and the verification module 5300 is further configured to pass the model verification on the acquired target domain prediction model if the value of the preset parameter of the acquired target domain prediction model is within a preset range.
In some embodiments, the obtained target domain prediction model is a solidified map file in which parameters of the target domain prediction model determined through training are converted into constants.
In some embodiments, the data to be tested of the user includes characteristics of the user and characteristics of the product, and the information pushing result for the corresponding user is a result of whether the product is recommended for the user.
An embodiment of the model training apparatus for information push of the present invention is described below with reference to fig. 6.
Fig. 6 shows a schematic structural diagram of a model training apparatus for information push according to some embodiments of the present invention. As shown in fig. 6, the model training apparatus 600 of this embodiment includes: a source domain training module 6100 configured to train the source domain prediction model with user training data of the source domain; the target domain training module 6200 is configured to input the source domain characteristic data and the target domain characteristic data corresponding to the same user into the source domain prediction model and the target domain prediction model respectively to obtain a source domain prediction result and a target domain prediction result corresponding to the user; and adjusting parameters of the target domain prediction model according to the difference between the source domain prediction result and the target domain prediction result and the difference between the marking value corresponding to the user and the target domain prediction result.
In some embodiments, the target domain training module 6200 is further configured to adjust parameters of the target domain prediction model based on a loss function of the target domain prediction model, wherein the loss function of the target domain prediction model comprises a cross entropy of the source domain prediction result and the target domain prediction result, and a cross entropy of the user corresponding token value and the target domain prediction result.
In some embodiments, the target domain prediction model is more complex than the source domain prediction model.
In some embodiments, the number of dimensions of the input data of the target domain prediction model is greater than the number of dimensions of the input data of the source domain prediction model.
In some embodiments, the source domain training module 6100 is further configured to obtain first and second feature data corresponding to each of a plurality of users, wherein the second feature data corresponding to the same user comprises features of a partial dimension of the first feature data; training a first preparatory model by using the first characteristic data; inputting first characteristic data and second characteristic data corresponding to the same user into the first preliminary model and the second preliminary model respectively to obtain a first preliminary model prediction result and a second preliminary model prediction result corresponding to the user; adjusting parameters of the second preparatory model according to the difference between the first preparatory model prediction result and the second preparatory model prediction result and the difference between the mark value corresponding to the user and the first preparatory model prediction result; and taking the adjusted second preparation model as the source domain prediction model.
In some embodiments, the first preparatory model and the second preparatory model have the same network model structure except that the input layers are different.
In some embodiments, the source domain prediction result and the target domain prediction result corresponding to the user are recommendation results associated with the same item.
An embodiment of the information push system of the present invention is described below with reference to fig. 7.
Fig. 7 illustrates a schematic structural diagram of an information push system according to some embodiments of the invention. As shown in fig. 7, the information push system 70 of this embodiment includes an information push apparatus 500 and a model training apparatus 600 for information push.
In some embodiments, the information pushing device 500 is deployed in an online system of the information pushing system 70, and the model training device 600 is deployed in an offline system of the information pushing system 70.
Fig. 8 shows a schematic structural diagram of a data processing device according to some embodiments of the present invention, the data processing device being an information pushing device or a model training device for information pushing. As shown in fig. 8, the data processing apparatus 80 of this embodiment includes: a memory 810 and a processor 820 coupled to the memory 810, the processor 820 being configured to execute the information pushing method or the model training method for information pushing in any of the aforementioned embodiments based on instructions stored in the memory 810.
Fig. 9 is a schematic structural diagram of a data processing apparatus according to another embodiment of the present invention, where the data processing apparatus is an information pushing apparatus or a model training apparatus for information pushing. As shown in fig. 9, the data processing apparatus 90 of this embodiment includes: the memory 910 and the processor 920 may further include an input/output interface 930, a network interface 940, a storage interface 950, and the like. These interfaces 930, 940, 950 and the memory 910 and the processor 920 may be connected, for example, by a bus 960. The input/output interface 930 provides a connection interface for input/output devices such as a display, a mouse, a keyboard, and a touch screen. The network interface 940 provides a connection interface for various networking devices. The storage interface 950 provides a connection interface for external storage devices such as an SD card and a usb disk.
An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program is configured to implement any one of the foregoing information pushing methods or a model training method for information pushing when executed by a processor.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (18)
1. An information push method, comprising:
acquiring a target domain prediction model which is sent by an offline system and completes training, wherein the target domain prediction model is obtained by utilizing a prediction result of a source domain prediction model and user training data of a target domain for training, and the source domain prediction model is obtained by utilizing the user training data of the source domain for training;
and predicting the user to-be-detected data of the target domain by using the target domain prediction model so as to obtain an information pushing result of the corresponding user.
2. The information push method according to claim 1, wherein a trained target domain prediction model and a coding value of the target domain prediction model transmitted by an offline system are periodically acquired, and the information push method further comprises:
under the condition that the code value of the obtained prediction network model is different from the code value of the currently used target domain prediction model, the version of the obtained target domain prediction model is verified;
and under the condition that the online verification of the obtained target domain prediction model passes, replacing the currently used target domain prediction model with the obtained target domain prediction model, wherein the online verification comprises the version verification.
3. The information push method according to claim 2, wherein the online verification further includes model verification, and the information push method further includes:
and under the condition that the value of the preset parameter of the obtained target domain prediction model is within a preset range, performing model verification on the obtained target domain prediction model.
4. The information pushing method according to claim 1, wherein the acquired target domain prediction model is a solidified map file in which parameters of the target domain prediction model determined by training are converted into constants.
5. The information pushing method according to claim 1, wherein the data to be tested of the user includes characteristics of the user and characteristics of a product, and the information pushing result for the corresponding user is a result of whether the product is recommended for the user.
6. A model training method for information push comprises the following steps:
training a source domain prediction model by using user training data of a source domain;
respectively inputting source domain characteristic data and target domain characteristic data corresponding to the same user into the source domain prediction model and the target domain prediction model to obtain a source domain prediction result and a target domain prediction result corresponding to the user;
and adjusting parameters of the target domain prediction model according to the difference between the source domain prediction result and the target domain prediction result and the difference between the marking value corresponding to the user and the target domain prediction result.
7. The model training method of claim 6, wherein said adjusting parameters of the target domain prediction model comprises:
and adjusting parameters of the target domain prediction model based on a loss function of the target domain prediction model, wherein the loss function of the target domain prediction model comprises the cross entropy of the source domain prediction result and the target domain prediction result and the cross entropy of the mark value corresponding to the user and the target domain prediction result.
8. The model training method of claim 6, wherein the complexity of the target domain prediction model is higher than the complexity of the source domain prediction model.
9. The model training method of claim 6, wherein the number of dimensions of the input data of the target domain prediction model is greater than the number of dimensions of the input data of the source domain prediction model.
10. The model training method of claim 6, wherein the training of the source domain predictive model using the source domain user training data comprises:
acquiring first characteristic data and second characteristic data corresponding to each of a plurality of users, wherein the second characteristic data corresponding to the same user comprises characteristics of partial dimensions of the first characteristic data;
training a first preparatory model by using the first characteristic data;
inputting first characteristic data and second characteristic data corresponding to the same user into the first preliminary model and the second preliminary model respectively to obtain a first preliminary model prediction result and a second preliminary model prediction result corresponding to the user;
adjusting parameters of the second preparatory model according to the difference between the first preparatory model prediction result and the second preparatory model prediction result and the difference between the mark value corresponding to the user and the first preparatory model prediction result;
and taking the adjusted second preparation model as the source domain prediction model.
11. The model training method of claim 10, wherein the first preliminary model and the second preliminary model have the same network model structure except that input layers are different.
12. The model training method of claim 6, wherein the source domain prediction result and the target domain prediction result corresponding to the user are recommendation results associated with the same item.
13. An information pushing apparatus comprising:
the system comprises an acquisition module, a storage module and a processing module, wherein the acquisition module is configured to acquire a target domain prediction model which is sent by an offline system and completes training, the target domain prediction model is acquired by utilizing a prediction result of a source domain prediction model and user training data of a target domain for training, and the source domain prediction model is acquired by utilizing user training data of the source domain for training;
and the prediction module is configured to predict the data to be tested of the user in the target domain by using the target domain prediction model so as to obtain an information push result of the corresponding user.
14. A model training device for information push, comprising:
a source domain training module configured to train a source domain prediction model using user training data of a source domain;
the target domain training module is configured to input source domain characteristic data and target domain characteristic data corresponding to the same user into the source domain prediction model and the target domain prediction model respectively to obtain a source domain prediction result and a target domain prediction result corresponding to the user; and adjusting parameters of the target domain prediction model according to the difference between the source domain prediction result and the target domain prediction result and the difference between the marking value corresponding to the user and the target domain prediction result.
15. An information push system, comprising:
the information push device of claim 13; and
the model training apparatus for information push of claim 14.
16. An information pushing apparatus comprising:
a memory; and
a processor coupled to the memory, the processor configured to perform the information pushing method of any of claims 1-5 based on instructions stored in the memory.
17. A model training device for information push, comprising:
a memory; and
a processor coupled to the memory, the processor configured to execute the model training method for information push of any one of claims 6-12 based on instructions stored in the memory.
18. A computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the information push method of any one of claims 1 to 5, or the model training method for information push of any one of claims 6 to 12.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011397968.4A CN112417293B (en) | 2020-12-03 | 2020-12-03 | Information pushing method and system, model training method and related equipment |
| PCT/CN2021/132104 WO2022116862A1 (en) | 2020-12-03 | 2021-11-22 | Information pushing method and system, model training method, and related devices |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011397968.4A CN112417293B (en) | 2020-12-03 | 2020-12-03 | Information pushing method and system, model training method and related equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112417293A true CN112417293A (en) | 2021-02-26 |
| CN112417293B CN112417293B (en) | 2024-06-18 |
Family
ID=74829847
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011397968.4A Active CN112417293B (en) | 2020-12-03 | 2020-12-03 | Information pushing method and system, model training method and related equipment |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN112417293B (en) |
| WO (1) | WO2022116862A1 (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022116862A1 (en) * | 2020-12-03 | 2022-06-09 | 京东科技控股股份有限公司 | Information pushing method and system, model training method, and related devices |
| CN116502271A (en) * | 2023-06-21 | 2023-07-28 | 杭州金智塔科技有限公司 | Privacy protection cross-domain recommendation method based on generation model |
| CN117609887A (en) * | 2024-01-19 | 2024-02-27 | 腾讯科技（深圳）有限公司 | Data enhancement model training and data processing method, device, equipment and medium |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109447149A (en) * | 2018-10-25 | 2019-03-08 | 腾讯科技（深圳）有限公司 | A kind of training method of detection model, device and terminal device |
| CN109544190A (en) * | 2018-11-28 | 2019-03-29 | 北京芯盾时代科技有限公司 | A kind of fraud identification model training method, fraud recognition methods and device |
| US20190205701A1 (en) * | 2017-12-29 | 2019-07-04 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method for Training Model and Information Recommendation System |
| CN110378726A (en) * | 2019-07-02 | 2019-10-25 | 阿里巴巴集团控股有限公司 | A kind of recommended method of target user, system and electronic equipment |
| CN110400169A (en) * | 2019-07-02 | 2019-11-01 | 阿里巴巴集团控股有限公司 | A kind of information-pushing method, device and equipment |
| CN110399856A (en) * | 2019-07-31 | 2019-11-01 | 上海商汤临港智能科技有限公司 | Feature extraction network training method, image processing method, device and its equipment |
| US20190354801A1 (en) * | 2018-05-16 | 2019-11-21 | Nec Laboratories America, Inc. | Unsupervised cross-domain distance metric adaptation with feature transfer network |
| CN110796232A (en) * | 2019-10-12 | 2020-02-14 | 腾讯科技（深圳）有限公司 | Attribute prediction model training method, attribute prediction method and electronic equipment |
| US20200104710A1 (en) * | 2018-09-27 | 2020-04-02 | Google Llc | Training machine learning models using adaptive transfer learning |
| CN111582449A (en) * | 2020-05-07 | 2020-08-25 | 广州视源电子科技股份有限公司 | Training method, device, equipment and storage medium for target domain detection network |
| CN111966914A (en) * | 2020-10-26 | 2020-11-20 | 腾讯科技（深圳）有限公司 | Content recommendation method and device based on artificial intelligence and computer equipment |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109544306B (en) * | 2018-11-30 | 2021-09-21 | 苏州大学 | Cross-domain recommendation method and device based on user behavior sequence characteristics |
| CN112417293B (en) * | 2020-12-03 | 2024-06-18 | 京东科技控股股份有限公司 | Information pushing method and system, model training method and related equipment |
-
2020
- 2020-12-03 CN CN202011397968.4A patent/CN112417293B/en active Active
-
2021
- 2021-11-22 WO PCT/CN2021/132104 patent/WO2022116862A1/en active Application Filing
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190205701A1 (en) * | 2017-12-29 | 2019-07-04 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method for Training Model and Information Recommendation System |
| US20190354801A1 (en) * | 2018-05-16 | 2019-11-21 | Nec Laboratories America, Inc. | Unsupervised cross-domain distance metric adaptation with feature transfer network |
| US20200104710A1 (en) * | 2018-09-27 | 2020-04-02 | Google Llc | Training machine learning models using adaptive transfer learning |
| CN109447149A (en) * | 2018-10-25 | 2019-03-08 | 腾讯科技（深圳）有限公司 | A kind of training method of detection model, device and terminal device |
| CN109544190A (en) * | 2018-11-28 | 2019-03-29 | 北京芯盾时代科技有限公司 | A kind of fraud identification model training method, fraud recognition methods and device |
| CN110378726A (en) * | 2019-07-02 | 2019-10-25 | 阿里巴巴集团控股有限公司 | A kind of recommended method of target user, system and electronic equipment |
| CN110400169A (en) * | 2019-07-02 | 2019-11-01 | 阿里巴巴集团控股有限公司 | A kind of information-pushing method, device and equipment |
| CN110399856A (en) * | 2019-07-31 | 2019-11-01 | 上海商汤临港智能科技有限公司 | Feature extraction network training method, image processing method, device and its equipment |
| CN110796232A (en) * | 2019-10-12 | 2020-02-14 | 腾讯科技（深圳）有限公司 | Attribute prediction model training method, attribute prediction method and electronic equipment |
| CN111582449A (en) * | 2020-05-07 | 2020-08-25 | 广州视源电子科技股份有限公司 | Training method, device, equipment and storage medium for target domain detection network |
| CN111966914A (en) * | 2020-10-26 | 2020-11-20 | 腾讯科技（深圳）有限公司 | Content recommendation method and device based on artificial intelligence and computer equipment |
Non-Patent Citations (1)
| Title |
|---|
| 程磊;高茂庭;: "基于深度神经网络的推荐算法", 现代计算机(专业版), no. 22 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022116862A1 (en) * | 2020-12-03 | 2022-06-09 | 京东科技控股股份有限公司 | Information pushing method and system, model training method, and related devices |
| CN116502271A (en) * | 2023-06-21 | 2023-07-28 | 杭州金智塔科技有限公司 | Privacy protection cross-domain recommendation method based on generation model |
| CN116502271B (en) * | 2023-06-21 | 2023-09-19 | 杭州金智塔科技有限公司 | Privacy protection cross-domain recommendation method based on generation model |
| CN117609887A (en) * | 2024-01-19 | 2024-02-27 | 腾讯科技（深圳）有限公司 | Data enhancement model training and data processing method, device, equipment and medium |
| CN117609887B (en) * | 2024-01-19 | 2024-05-10 | 腾讯科技（深圳）有限公司 | Data enhancement model training and data processing method, device, equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112417293B (en) | 2024-06-18 |
| WO2022116862A1 (en) | 2022-06-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112417293A (en) | Information pushing method and system, model training method and related equipment | |
| CN108509975B (en) | Abnormal online clustering method and device and electronic equipment | |
| CN114780831A (en) | Sequence recommendation method and system based on Transformer | |
| CN112734104B (en) | Cross-domain recommendation method fusing generation countermeasure network and self-encoder | |
| CN112835682B (en) | Data processing method, device, computer equipment and readable storage medium | |
| CN113449011A (en) | Big data prediction-based information push updating method and big data prediction system | |
| CN112860676B (en) | Data cleaning method applied to big data mining and business analysis and cloud server | |
| CN111460290A (en) | Information recommendation method, device, equipment and storage medium | |
| CN106778851A (en) | Social networks forecasting system and its method based on Mobile Phone Forensics data | |
| CN115114329A (en) | Method and device for detecting data stream abnormity, electronic equipment and storage medium | |
| CN111382403A (en) | Training method, device, equipment and storage medium of user behavior recognition model | |
| CN113516251B (en) | Machine learning system and model training method | |
| CN113449753A (en) | Service risk prediction method, device and system | |
| CN109710890B (en) | Method and system for identifying false material in real time based on constructed behavior portrait model | |
| CN111162918B (en) | Service fulfillment management method and device | |
| CN111833074A (en) | Dispute responsibility confirmation method and device for network appointment vehicle and computer readable storage medium | |
| CN111709231B (en) | Class case recommendation method based on self-attention variational self-coding | |
| CN113140012A (en) | Image processing method, image processing apparatus, image processing medium, and electronic device | |
| CN112529218A (en) | Building safety detection method and system based on correlation analysis | |
| CN109190649B (en) | Optimization method and device for deep learning network model server | |
| CN111651500A (en) | User identity recognition method, electronic device and storage medium | |
| CN116468525A (en) | Information processing method, information processing device, electronic equipment and computer storage medium | |
| CN116578665A (en) | Method and equipment for jointly extracting extensible text information based on prompt learning | |
| CN112328779B (en) | Training sample construction method, device, terminal equipment and storage medium | |
| CN113657613A (en) | Prediction model training method, data processing device and system |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: Room 221, 2 / F, block C, 18 Kechuang 11th Street, Daxing District, Beijing, 100176 Applicant after: Jingdong Technology Holding Co.,Ltd. Address before: Room 221, 2 / F, block C, 18 Kechuang 11th Street, Daxing District, Beijing, 100176 Applicant before: Jingdong Digital Technology Holding Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |