CN111428783A - Method and device for performing sample domain conversion on training samples of recommendation model - Google Patents
Method and device for performing sample domain conversion on training samples of recommendation model Download PDFInfo
- Publication number
- CN111428783A CN111428783A CN202010205866.1A CN202010205866A CN111428783A CN 111428783 A CN111428783 A CN 111428783A CN 202010205866 A CN202010205866 A CN 202010205866A CN 111428783 A CN111428783 A CN 111428783A
- Authority
- CN
- China
- Prior art keywords
- sample
- training sample
- training
- migration
- domain
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Abstract
The embodiment of the specification provides a method and a device for performing sample domain conversion on training samples. And determining the third training sample as the training sample in the first sample domain when the first constraint condition is met.
Description
Technical Field
The embodiment of the specification relates to the technical field of machine learning, in particular to a method and a device for performing sample domain conversion on training samples of a recommendation model.
Background
The samples in each scene can reflect the service characteristics of the scene, and based on the service characteristics, the samples in different scenes are different. For example, each business mode is a scene, and different business modes have different business characteristics, for example, different consumer groups and different types of goods are targeted, so that samples required by cold start of different business modes are different, for example, a sample in a scene of the "mobile phone section" is mainly a mobile phone, and a sample in a scene of the "double eleven" is mainly a garment, and the like.
For a mature scene, the number of samples is rich. For a new scene, the corresponding number of samples is small, and even the number of samples does not meet the requirement of cold start. At this time, the samples in the new scene need to be added, and one way to add the samples in the new scene is to use the samples in the mature scene.
Disclosure of Invention
In view of the foregoing, embodiments of the present specification provide a method and an apparatus for performing sample domain conversion on training samples of a recommendation model. In the method, after a first training sample in a first sample domain and a second training sample in a second sample domain are obtained, the second training sample is migrated and converted into a third training sample in the first sample domain by using a first sample migration and conversion process, wherein the first sample migration and conversion process is realized under a first constraint condition by using a machine learning method. By the method, the third training sample can be suitable for the scene of the first sample domain under the first constraint condition, so that the training samples in the first sample domain are expanded, and the corresponding recommendation model can be trained by using the expanded training samples in the first sample domain.
According to an aspect of an embodiment of the present specification, there is provided a method of performing sample domain conversion on training samples of a recommendation model, including: acquiring a first training sample under a first sample domain and a second training sample under a second sample domain; wherein the training samples in the first sample domain are used for training a first recommendation model for recommending a first class of commodities, the training samples in the second sample domain are used for training a second recommendation model for recommending a second class of commodities, the first training samples comprise samples for the first class of commodities, and the second training samples comprise samples for the second class of commodities; migrating the second training sample into a third training sample using a first sample migration transformation process; and determining the third training sample as a training sample in the first sample domain such that the first recommendation model is trained based on the determined training sample, wherein the first sample migration transformation process is implemented using a machine learning method under a first constraint comprising: and the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a first threshold value, wherein the distributed distance of each migration conversion is the distributed distance between the third training sample obtained by the migration conversion and the first training sample.
Optionally, in an example of the above aspect, the first training sample and the second training sample are supervised training samples, and the first constraint further includes: the training error of the third training sample is smaller than a second threshold value, and the prediction error of the first training sample is smaller than a third threshold value, wherein the prediction value of the first training sample is predicted by using a first prediction model, and the first prediction model is trained by using the third training sample.
Optionally, in an example of the above aspect, further comprising: using a second sample migration transformation process to transform the third training sample migration into a fourth training sample, wherein the second sample migration transformation process is implemented using a machine learning method under a second constraint; and determining the third training sample as a training sample under the first sample domain comprises: determining the third training sample as the training sample under the first sample domain when the fourth training sample satisfies a third constraint condition, wherein the third constraint condition includes: a distributed distance between the fourth training sample and the second training sample is less than a fourth threshold.
Optionally, in an example of the above aspect, the second training sample and the fourth training sample are supervised training samples, and the third constraint further includes: the training error of the fourth training sample is smaller than a fifth threshold value, and the prediction error of the second training sample is smaller than a sixth threshold value, wherein the predicted value of the second training sample is predicted by using a second prediction model, and the second prediction model is trained by using the fourth training sample.
Optionally, in an example of the above aspect, further comprising: and when the fourth training sample does not meet a third constraint condition, determining the fourth training sample as a second training sample under a second sample domain, and returning to execute the operation of converting the second training sample into a third training sample by using the first sample migration conversion process.
Optionally, in an example of the above aspect, the second constraint includes: and the cycle number or the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a seventh threshold, wherein the distributed distance of each migration conversion is the distributed distance between the fourth training sample and the second training sample obtained by the migration conversion.
Optionally, in one example of the above aspect, the first sample domain is a sample domain corresponding to a few-sample scene and the second sample domain is a sample domain corresponding to a multiple-sample scene.
Optionally, in one example of the above aspect, the distributed distance comprises at least one of a Euclidean distance, a K L divergence, a JS divergence, and an EM distance.
According to another aspect of the embodiments of the present specification, there is also provided an apparatus for performing sample domain conversion on training samples of a recommendation model, including: a training sample obtaining unit obtains a first training sample in a first sample domain and a second training sample in a second sample domain; wherein the training samples in the first sample domain are used for training a first recommendation model for recommending a first class of commodities, the training samples in the second sample domain are used for training a second recommendation model for recommending a second class of commodities, the first training samples comprise samples for the first class of commodities, and the second training samples comprise samples for the second class of commodities; a first sample migration conversion unit migration-converts the second training sample into a third training sample using a first sample migration conversion process; and a first training sample determination unit determining the third training sample as a training sample in the first sample domain, such that the first recommendation model is trained based on the determined training sample, wherein the first sample migration transition process is implemented using a machine learning method under a first constraint condition, the first constraint condition including: and the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a first threshold value, wherein the distributed distance of each migration conversion is the distributed distance between the third training sample obtained by the migration conversion and the first training sample.
Optionally, in an example of the above aspect, the first training sample and the second training sample are supervised training samples, and the first constraint further includes: the training error of the third training sample is smaller than a second threshold value, and the prediction error of the first training sample is smaller than a third threshold value, wherein the prediction value of the first training sample is predicted by using a first prediction model, and the first prediction model is trained by using the third training sample.
Optionally, in an example of the above aspect, further comprising: a second sample migration conversion unit migration-converts the third training sample into a fourth training sample using a second sample migration conversion process, wherein the second sample migration conversion process is implemented under a second constraint condition using a machine learning method; and the first training sample determination unit: determining the third training sample as the training sample under the first sample domain when the fourth training sample satisfies a third constraint condition, wherein the third constraint condition includes: a distributed distance between the fourth training sample and the second training sample is less than a fourth threshold.
Optionally, in an example of the above aspect, the second training sample and the fourth training sample are supervised training samples, and the third constraint further includes: the training error of the fourth training sample is smaller than a fifth threshold value, and the prediction error of the second training sample is smaller than a sixth threshold value, wherein the predicted value of the second training sample is predicted by using a second prediction model, and the second prediction model is trained by using the fourth training sample.
Optionally, in an example of the above aspect, further comprising: and the second training sample determining unit is used for determining the fourth training sample as a second training sample in a second sample domain when the fourth training sample does not meet a third constraint condition, and triggering the first sample migration and conversion unit to execute the operation of migrating and converting the second training sample into a third training sample by using the first sample migration and conversion process.
Optionally, in an example of the above aspect, the second constraint includes: and the cycle number or the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a seventh threshold, wherein the distributed distance of each migration conversion is the distributed distance between the fourth training sample and the second training sample obtained by the migration conversion.
According to another aspect of embodiments herein, there is also provided an electronic device, including: at least one processor; and a memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform the method for sample domain conversion of training samples as described above.
According to another aspect of embodiments herein, there is also provided a machine-readable storage medium storing executable instructions that, when executed, cause the machine to perform the method for sample domain conversion of training samples as described above.
Drawings
A further understanding of the nature and advantages of the present disclosure may be realized by reference to the following drawings. In the drawings, similar components or features may have the same reference numerals. The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the detailed description serve to explain the embodiments of the invention. In the drawings:
FIG. 1 shows a flow diagram of a method for sample domain conversion of training samples of an embodiment of the present description;
FIG. 2 illustrates a flow diagram of one example of a first sample migration translation process of an embodiment of the present specification;
FIG. 3 shows a flow diagram of a method for sample domain conversion of training samples of another embodiment of the present description;
FIG. 4 illustrates a flow diagram of one example of a second sample migration translation process of another embodiment of the present description;
FIG. 5 illustrates a block diagram of an apparatus for sample domain conversion of training samples in accordance with an embodiment of the present specification;
FIG. 6 shows a block diagram of an apparatus for sample domain conversion of training samples according to another embodiment of the present description; and
FIG. 7 illustrates a block diagram of an electronic device for sample domain conversion of training samples in accordance with an embodiment of the present description.
Detailed Description
The subject matter described herein will be discussed with reference to example embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and thereby implement the subject matter described herein, and are not intended to limit the scope, applicability, or examples set forth in the claims. Changes may be made in the function and arrangement of elements discussed without departing from the scope of the disclosure. Various examples may omit, substitute, or add various procedures or components as needed. In addition, features described with respect to some examples may also be combined in other examples.
As used herein, the term "include" and its variants mean open-ended terms in the sense of "including, but not limited to. The term "based on" means "based at least in part on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". The terms "first," "second," and the like may refer to different or the same object. Other definitions, whether explicit or implicit, may be included below. The definition of a term is consistent throughout the specification unless the context clearly dictates otherwise.
The samples in each scene can reflect the service characteristics of the scene, and based on the service characteristics, the samples in different scenes are different. For example, each business mode is a scene, and different business modes have different business characteristics, for example, different consumer groups and different types of goods are targeted, so that samples required by cold start of different business modes are different, for example, a sample in a scene of the "mobile phone section" is mainly a mobile phone, and a sample in a scene of the "double eleven" is mainly a garment, and the like.
For a mature scene, the number of samples is rich. For a new scene, the corresponding number of samples is small, and even the number of samples does not meet the requirement of cold start. At this point, samples in the new scene need to be added. Currently, one way to increase this is to rely on manual experience, i.e. on experts with a rich product experience to choose samples that are suitable for the new scene. Another way to add is to directly take the sample in the mature scene as the sample in the new scene.
However, the first approach relies on manual experience, resulting in waste of human resources, and the manual experience cannot be quantitatively evaluated, resulting in that the selected samples are not suitable for new scenes. In the second method, the samples in the mature scene are not suitable for the new scene because the service characteristics of the new scene and the mature scene are different.
In order to solve the above problem, embodiments of the present specification provide a method and an apparatus for performing sample domain conversion on training samples of a recommendation model. In the method, after a first training sample in a first sample domain and a second training sample in a second sample domain are obtained, the second training sample is migrated and converted into a third training sample in the first sample domain by using a first sample migration and conversion process, wherein the first sample migration and conversion process is realized under a first constraint condition by using a machine learning method. By the method, the training samples in the second sample domain are migrated and converted into the third training samples in the first sample domain, the third training samples can be suitable for the scene of the first sample domain under the first constraint condition, so that the training samples in the first sample domain are expanded, and the corresponding recommendation model can be trained by using the expanded training samples in the first sample domain.
A method and an apparatus for performing sample domain conversion on training samples of a recommendation model according to an embodiment of the present specification will be described in detail below with reference to the accompanying drawings.
FIG. 1 illustrates a flow diagram of a method of sample domain conversion of training samples of a recommendation model of an embodiment of the present specification.
As shown in FIG. 1, at block 110, a first training sample in a first sample domain and a second training sample in a second sample domain are obtained.
In the embodiment of the present specification, the first sample field and the second sample field include a plurality of training samples, and the training samples belonging to the same sample field are for the same type of goods. The first training samples comprise samples for a first class of goods and the second training samples comprise samples for a second class of goods. Wherein the first type of merchandise and the second type of merchandise comprise different merchandise. In one example, the first type of item and the second type of item may comprise the same item. For example, the application scenes of the twenty-one correspond to the first category of commodities, and the application scenes of the 3C digital sections correspond to the second category of commodities, because the twenty-one is for the full category of commodities, and the 3C digital sections are only for the 3C digital products, the second category of commodities and the first category of commodities both include the 3C digital products.
In the embodiment of the present specification, the training samples in the first sample domain are used for training a first recommendation model for recommending a first type of commodity, and the training samples in the second sample domain are used for training a second recommendation model for recommending a second type of commodity.
The first sample field corresponds to at least one application scenario (e.g., a first application scenario) and the second sample field corresponds to at least one application scenario (e.g., a second application scenario). The first recommendation model is applied to a first application scenario and the second recommendation model is applied to a second application scenario.
For example, the first application scenario corresponding to the first sample domain is a "men's clothing festival", the first type of goods in the scenario are men's clothing, and the training samples (e.g., the first training samples) in the first sample domain are all for men's clothing, such as men's clothing of various types, various styles and various brands. The second application scene corresponding to the second sample domain is a "mobile phone festival" scene, the second type of goods in the scene are mobile phones, and the second training samples in the second sample domain are all specific to the mobile phones, such as various brands, various functions, various types of mobile phones, and the like.
In the embodiment of the present specification, the training samples may be represented by variables of different dimensions, and thus, each training sample may be composed of variables of multiple dimensions, such as commodity sales volume, return rate, price, and unrevowed conversion rate. The variables of the first training sample have the same variables as the variables of the second training sample. In one example, the variables of the first training sample are the same as the variables of the second training sample, e.g., the variables of the first training sample and the second training sample are both comprised of commodity sales, return rate, price, and unrevowed conversion rate. Each training sample may be represented as: (X)1，X2，……，Xn) Wherein X is1，X2，……，XnRepresenting n feature variables, n may be specified.
In one example of the embodiments of the present specification, the first sample domain is a sample domain corresponding to a few-sample scene, and the second sample domain is a sample domain corresponding to a multiple-sample scene. The number of samples in the sample domain corresponding to the few-sample scene is small, and even the samples are not enough to be used as training samples for training the prediction model of the few-sample scene; the number of samples in the sample domain corresponding to the multi-sample scene is large.
In one example, a few sample scene may be a new scene and a many sample scene may be a mature scene that has been applied multiple times. For example, a mature scenario is a "dueleven" promotional program that has been maturely applied, while a new scenario is another promotional program that has not yet been promoted.
By the above example, a plurality of training samples correspond to a plurality of sample scenes, and the training samples in the sample domain corresponding to the small sample scene are increased by using the training samples in the plurality of sample scenes, so that when the prediction model in the small sample scene needs to be trained, the sample domain corresponding to the small sample scene can provide a sufficient number of training samples.
At block 120, the second training sample is migration converted to a third training sample using a first sample migration conversion process.
In an embodiment of the present specification, the first sample migration transformation process is implemented under the first constraint using a machine learning method in which the second training sample is an input and the third training sample is an output.
FIG. 2 illustrates a flow diagram of one example of a first sample migration translation process of an embodiment of the present specification. As shown in fig. 2, in block 121, the second training sample is migrated and converted into a third training sample, after the third training sample is obtained through each conversion, in block 123, it is determined whether the first constraint condition is currently satisfied, if so, the process proceeds to block 125, and the first sample migration and conversion process is ended. If not, the process proceeds to block 127, the parameters of the machine learning model are adjusted, and the process returns to block 121, that is, the second training sample is migrated to the third training sample using the adjusted machine learning model.
At block 127, parameter adjustment is performed based on the current parameters of the machine learning model, and the third training sample converted by the adjusted machine learning model is enabled to satisfy the first constraint condition, or more closely satisfy the first constraint condition. In one example, the machine learning model may be a first conversion function that may be used to transition the second training sample migration to the third training sample.
In the embodiment of the present specification, the first sample migration transition process is a loop process, and the control condition of the loop is the first constraint condition. In each cycle, a third training sample corresponding to the time can be obtained through conversion, and a distributed distance of the first sample migration conversion can be obtained based on the third training sample, that is, the distributed distance between the third training sample and the first training sample, and the distributed distance can be expressed by the following formula:
|S′(X′1，X′2，…，X′n)，T(X1，X2，…，Xn)|
wherein S 'represents a third training sample, X'1，X′2，…，X′nA characteristic variable representing a third training sample, T representing the first training sample, X1，X2，…，XnRepresenting the feature variable of the first training sample, the symbol "|" represents the inter-distribution distance.
The distributed distance between the third training sample and the first training sample can represent the distribution difference between the data of the third training sample and the data of the first training sample, and the larger the distributed distance is, the larger the distribution difference between the data is, the worse the migration conversion effect is; the smaller the distribution distance is, the smaller the distribution difference among the data is, and the better the migration conversion effect is. When the distribution distance is 0, it means that the third training sample is the same as the first training sample.
In one example of an embodiment of the present specification, the distributed distances may include at least one of Euclidean distances, K L divergence (Kullback-L eibler divergence), JS divergence (Jensen-Shannon divergence), and EM (Earth-Mover) distances.
There are M first training samples under the first sample domain, and the feature variable dimension of each training sample is n, then the M first training samples can be expressed as:
wherein, P1The representation includes M first trainingsMatrix of training samples, ajiRepresenting the ith feature variable in the jth first training sample.
The number of third training samples obtained by migration and conversion is N, and the feature variable dimension of each training sample is N, so that the N third training samples can be expressed as:
wherein, P2Representing a matrix comprising N third training samples, bjiRepresenting the ith feature variable in the jth third training sample.
The euclidean distances between the N third training samples and the M first training samples are:
wherein, | | Aj||2＝aj1 2+…+aji 2+…+ajn 2，||Bj||2＝bj1 2+…+bji 2+…+bjn 2。P2 TRepresents P2The transposed matrix of (2).
In the embodiment of the present specification, the first constraint condition includes: the difference between the distributed distance of the current first sample migration transition and the distributed distance of the last first sample migration transition is less than a first threshold. Wherein the first threshold may be specified.
For example, if the current migration conversion is a third migration conversion process circulating in the first sample migration conversion process, the distributed distance of the third migration conversion may be calculated, and then the difference between the distributed distance of the third migration conversion and the distributed distance of the second migration conversion may be calculated, and it is determined whether the difference is smaller than the first threshold.
Based on the first constraint condition, the distributed distance obtained in each migration and conversion process is smaller than the distributed distance obtained in the last migration and conversion process, and the difference between the two adjacent distributed distances is gradually reduced, so that the effect of migrating and converting the second training sample into the third training sample is better, and the third training sample obtained by migration and conversion is closer to the first training sample in the first sample domain.
At block 130, a third training sample is determined as the training sample in the first sample domain.
In an embodiment of the present specification, the determined third training sample is obtained by migrating the transition when the first constraint condition is satisfied. For example, if the third training sample obtained by the conversion is migrated at the second time and satisfies the first constraint condition, the third training sample is determined as the training sample in the first sample domain.
In an example of the embodiment of the present specification, the third training sample and the first training sample are training samples for a first type of commodity, and may be used to train a first recommendation model for recommending the first type of commodity. In another example, the first recommendation model for recommending the first category of goods may also be trained with only the third training sample. For example, when the first recommendation model is applied to a scene of a "mobile phone section", a mobile phone can be recommended to a consumer through the trained first recommendation model.
In one example of an embodiment of the present specification, the first training sample and the second training sample are supervised training samples. The variables in each training sample include feature variables and label variables, and the label variables serve as labels in supervised training.
The model used for training by the supervised training sample can be a prediction model, wherein the input of the prediction model is a characteristic variable, the output of the prediction model is a predicted value corresponding to a label variable, and the predicted value is compared with the label variable in the training sample. Each training sample may be represented as: (X)1，X2，…，Xn(ii) a Y), wherein X1，X2，…，XnN feature variables are represented, and Y represents a tag variable. Tag variables are specifiable.
For example, each training sample is composed of a commodity sales volume, a return rate, a price and a non-return conversion rate, wherein the commodity sales volume, the return rate and the price are characteristic variables, and the non-return conversion rate is a label variable, so that the prediction model trained by the training samples can predict the non-return conversion rate of the commodity according to the commodity sales volume, the return rate and the price of the commodity.
The first constraint may further include: the training error of the third training sample is less than the second threshold, and the prediction error of the first training sample is less than the third threshold. Wherein both the second threshold and the third threshold may be specified.
The third training sample is obtained by the migration and conversion of the second training sample, and the third training sample is also a supervised training sample. And training by using the third training sample to obtain a first prediction model, wherein the first prediction model can predict the predicted value of the label variable according to the characteristic variable. For example, the first predictive model may use y-g1(x) Where x represents the input of the first prediction model, and y represents the output of the first prediction model (i.e. the predicted value corresponding to the tag variable), and further, the first prediction model can be further represented as: label variable g1(characteristic variables).
The training error of the third training sample is the training error for the first prediction model, and can be expressed as: | Y3-g1(S′(X′1，X′2，…，X′n) In which Y is a hydrogen atom, in which3Representing the value, X ', corresponding to the tag variable in the third training sample'1，X′2，…，X′nRepresents a feature variable, S ' (X ') in the third training sample '1，X′2，…，X′n) Representing a third training sample.
After the first prediction model is trained using the third training samples, the first prediction model may be used to predict the predicted values of the first training samples. The predicted value of the first training sample may be expressed as: g1(T(X1，X2，…，Xn) Wherein, T (X)1，X2，…，Xn) Represents the first training sample, X1，X2，…，XnRepresenting feature variables of the first training sample.
The prediction error of the first training sample is the error of the first prediction model predicting the first training sample, and can be expressed as: | Y1-g(T(X1，X2，…，Xn) In which Y is a hydrogen atom, in which1Representing the values corresponding to the feature variables in the first training sample.
In this example, satisfying the first constraint requires that the following three conditions be satisfied simultaneously: the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a first threshold value, the training error of the third training sample is smaller than a second threshold value, and the prediction error of the first training sample is smaller than a third threshold value.
Through the above example, the prediction error of the first training sample may reflect the migration conversion effect of the first sample migration conversion process, and if the migration conversion effect is better, that is, the closer the third training sample is to the first training sample, the smaller the prediction error of the first training sample predicted by using the first prediction model trained by the third training sample is. Therefore, the prediction error of the first training sample is used as one of the first constraints, so that the migration conversion effect of the first sample migration conversion can be improved, and the third training sample is closer to the first training sample.
FIG. 3 illustrates a flow diagram of a method for sample domain conversion of training samples of another embodiment of the present description. The embodiment shown in fig. 3 is a modified embodiment of the embodiment shown in fig. 1 described above.
In the embodiment shown in fig. 3, the operations of blocks 310, 320, and 350 are the same as blocks 110, 120, and 130, respectively, in fig. 1 and will not be described again. Only the differences will be described in detail below.
Following block 320, at block 330, the third training sample is then migrated to a fourth training sample using a second sample migration transformation process.
In an embodiment of the present specification, the second sample migration transition process is implemented using a machine learning method under the second constraint. In the machine learning method, the third training sample is input, and the fourth training sample is output.
Fig. 4 shows a flowchart of one example of a second sample migration conversion process of an embodiment of the present specification. As shown in fig. 4, at block 331, the third training sample is migrated and converted into a fourth training sample, and after the fourth training sample is obtained through each conversion, at block 333, it is determined whether the second constraint condition is currently satisfied, and if so, the process proceeds to block 335, and the second sample migration and conversion process is ended. If not, the process proceeds to block 337, the parameters of the machine learning model are adjusted, and the process returns to block 331, i.e., the third training sample is migrated to the fourth training sample using the adjusted machine learning model.
At block 337, parameter adjustment is performed based on the current parameters of the machine learning model, and the fourth training sample converted by the adjusted machine learning model is enabled to satisfy the second constraint condition, or more closely satisfy the second constraint condition. In one example, the machine learning model may be a second transfer function that may be used to transfer the third training sample into a fourth training sample.
In one example of an embodiment of the present specification, the first sample migration transformation process and the second sample migration transformation process may be implemented with a CYC L E-GAN model.
In one example of the embodiments of the present specification, the second constraint may include: the number of cycles, or the difference between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion, is less than a seventh threshold.
In this example, the number of cycles is the number of times of transition of the third training sample to the fourth training sample in the second sample transition process, and the number of cycles may be specified.
In this example, the seventh threshold may be specified. The second sample shift conversion process is a loop process, and the control condition of the loop is a second constraint condition. In each cycle, the distributed distance of each migration conversion is the distributed distance between the fourth training sample and the second training sample obtained by the migration conversion, and can be represented as:
|S″(X″1，X″2，…，X″n)，S(X1，X2，…，Xn)|
wherein S "represents a fourth training sample, X ″1，X″2，…，X″nThe feature variables of the fourth training sample are represented, and S represents the second training sample.
The distributed distance between the fourth training sample and the second training sample can represent the distribution difference between the data of the fourth training sample and the data of the second training sample, and the larger the distributed distance is, the worse the migration conversion effect of the first sample migration conversion process and the second sample migration conversion process is; the smaller the distribution distance is, the better the migration conversion effect of the first sample migration conversion process and the second sample migration conversion process is. When the fourth training sample is the same as the second training sample, the distributed distance between the fourth training sample and the second training sample is 0, in this case, the first sample migration conversion process migrates and converts the second training sample into the third training sample, the second sample migration conversion process migrates and converts the third training sample back into the second training sample, the migration conversion effects of the first sample migration conversion process and the second sample migration conversion process are good, and information loss is not caused.
At block 340, it is determined whether the fourth training sample satisfies the third constraint, and if so, flow proceeds to block 350 to determine the third training sample as the training sample in the first sample domain. If not, flow proceeds to block 360.
In an embodiment of the present specification, the third constraint includes: the distributed distance between the fourth training sample and the second training sample is less than a fourth threshold. Wherein the fourth threshold may be specified.
At block 360, a fourth training sample is determined as the second training sample in the second sample domain. And flow returns to block 320 to migrate the new second training sample to the third training sample using the first sample migration transformation process.
In one example of an embodiment of the present specification, the second training sample and the fourth training sample are supervised training samples. The variables in each training sample include feature variables and label variables.
The third constraint may further include: the training error of the fourth training sample is less than a fifth threshold and the prediction error of the second training sample is less than a sixth threshold. Wherein, the fifth threshold and the sixth threshold may be both specified.
And training a second prediction model by using the fourth training sample, wherein the second prediction model can predict corresponding label variables according to the characteristic variables. The second predictive model may use y-g2(x) X represents the input to the second prediction model and y represents the output of the second prediction model.
The training error of the fourth training sample is the training error for the second prediction model, and can be expressed as: | Y4-g2(S″(X″1，X″2，…，X″n) In which Y is a hydrogen atom, in which4Indicates the corresponding value of the tag variable, S '(X') in the fourth training sample1，X″2，…，X″n) A fourth training sample is represented.
After the second prediction model is trained using the fourth training sample, the second prediction model may be used to predict a predicted value of the second training sample. The predicted value of the second training sample may be expressed as: g2(S(X1，X2，…，Xn) Wherein, S (X)1，X2，…，Xn) Representing a second training sample.
The prediction error of the second training sample is an error of the second prediction model predicting the second training sample, and can be expressed as: | Y2-g2(S(X1，X2，…，Xn) In which Y is a hydrogen atom, in which2Representing the value of the tag variable in the second training sample.
In this example, satisfying the third constraint requires that the following three conditions be satisfied simultaneously: the distributed distance between the fourth training sample and the second training sample is less than a fourth threshold, the training error of the fourth training sample is less than a fifth threshold, and the prediction error of the second training sample is less than a sixth threshold.
By the above example, the prediction error of the fourth training sample may reflect the migration conversion effect of the second sample migration conversion process, and further may reflect the migration conversion effect of the first sample migration conversion process. If the migration conversion effect in the first sample migration conversion process is better, no information is lost in the process of migrating and converting the second training sample into the third training sample, and no information is lost in the process of migrating and converting the third training sample into the fourth training sample in the second sample migration conversion process, the fourth training sample is the same as the second training sample. If the distributed distance between the fourth training sample and the second training sample is larger, the larger the training error of the fourth training sample and the prediction error of the second training sample are, the serious information loss in the migration conversion process is indicated, and the worse migration conversion effect is.
Fig. 5 shows a block diagram of an apparatus for performing sample domain conversion on training samples (hereinafter referred to as a sample domain conversion apparatus 500) according to an embodiment of the present specification. As shown in fig. 5, the sample domain converting apparatus 500 includes a training sample acquiring unit 510, a first sample migration converting unit 520, and a first training sample determining unit 530.
The training sample acquiring unit 510 acquires a first training sample in a first sample domain and a second training sample in a second sample domain.
The first sample migration conversion unit 520 migration-converts the second training sample into a third training sample using the first sample migration conversion process. Wherein the first sample migration transformation process is implemented using a machine learning method under a first constraint comprising: and the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a first threshold value, and the distributed distance of each migration conversion is the distributed distance between the third training sample and the first training sample obtained by the migration conversion.
The first training sample determination unit 530 determines the third training sample as a training sample in the first sample domain.
In one example of an embodiment of the present specification, the first training sample and the second training sample are supervised training samples, and the first constraint may further include: and the training error of the third training sample is smaller than the second threshold, and the prediction error of the first training sample is smaller than the third threshold, wherein the prediction value of the first training sample is predicted by using the first prediction model, and the first prediction model is trained by using the third training sample.
Fig. 6 shows a block diagram of an apparatus for sample domain conversion of training samples according to another embodiment of the present specification. The embodiment shown in fig. 6 is a modified embodiment of the embodiment shown in fig. 5 described above.
As shown in fig. 6, the sample domain converting apparatus 500 may further include a second sample migration converting unit 540, the second sample migration converting unit 540 migration-converts the third training sample into the fourth training sample using a second sample migration converting process, wherein the second sample migration converting process is implemented using a machine learning method under a second constraint condition; and the first training sample determining unit 530 determines the third training sample as the training sample in the first sample domain when the fourth training sample satisfies a third constraint condition, where the third constraint condition includes: the distributed distance between the fourth training sample and the second training sample is less than a fourth threshold.
In one example, the second training sample and the fourth training sample are supervised training samples, and the third constraint further comprises: and the training error of the fourth training sample is smaller than a fifth threshold value, and the prediction error of the second training sample is smaller than a sixth threshold value, wherein the predicted value of the second training sample is predicted by using a second prediction model, and the second prediction model is trained by using the fourth training sample.
In one example, the sample domain converting apparatus 500 may further include a second training sample determining unit 550. When the fourth training sample does not satisfy the third constraint condition, the second training sample determination unit 550 determines the fourth training sample as the second training sample in the second sample domain, and triggers the first sample migration conversion unit to perform an operation of migrating and converting the second training sample into the third training sample using the first sample migration conversion process.
In one example, the second constraint includes: and the cycle number or the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a seventh threshold, wherein the distributed distance of each migration conversion is the distributed distance between the fourth training sample and the second training sample obtained by the migration conversion.
Embodiments of a method and apparatus for sample domain conversion of training samples according to embodiments of the present specification are described above with reference to fig. 1 to 6.
The apparatus for performing sample domain conversion on training samples according to the embodiments of the present disclosure may be implemented by hardware, or may be implemented by software, or a combination of hardware and software. The software implementation is taken as an example, and is formed by reading corresponding computer program instructions in the storage into the memory for operation through the processor of the device where the software implementation is located as a logical means. In the embodiments of the present specification, the apparatus for performing sample domain conversion on training samples may be implemented by an electronic device, for example.
Fig. 7 illustrates a block diagram of an electronic device 700 for sample domain conversion of training samples in accordance with an embodiment of the present description.
As shown in fig. 7, electronic device 700 may include at least one processor 710, storage (e.g., non-volatile storage) 720, memory 730, and communication interface 740, and at least one processor 710, storage 720, memory 730, and communication interface 740 are connected together via a bus 750. The at least one processor 710 executes at least one computer-readable instruction (i.e., the elements described above as being implemented in software) stored or encoded in memory.
In one embodiment, computer-executable instructions are stored in the memory that, when executed, cause the at least one processor 710 to: obtaining a first training sample in a first sample domain and a second training sample in a second sample domain, using a first sample migration conversion process to convert the second training sample into a third training sample, and determining the third training sample as the training sample in the first sample domain, wherein the first sample migration conversion process is implemented under a first constraint condition by using a machine learning method, and the first constraint condition comprises: and the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a first threshold value, and the distributed distance of each migration conversion is the distributed distance between the third training sample and the first training sample obtained by the migration conversion.
It should be appreciated that the computer-executable instructions stored in the memory, when executed, cause the at least one processor 710 to perform the various operations and functions described above in connection with fig. 1-6 in the various embodiments of the present description.
According to one embodiment, a program product, such as a machine-readable medium, is provided. A machine-readable medium may have instructions (i.e., elements described above as being implemented in software) that, when executed by a machine, cause the machine to perform various operations and functions described above in connection with fig. 1-6 in the various embodiments of the present specification.
Specifically, a system or apparatus may be provided which is provided with a readable storage medium on which software program code implementing the functions of any of the above embodiments is stored, and causes a computer or processor of the system or apparatus to read out and execute instructions stored in the readable storage medium.
In this case, the program code itself read from the readable medium can realize the functions of any of the above-described embodiments, and thus the machine-readable code and the readable storage medium storing the machine-readable code form part of the present invention.
Examples of the readable storage medium include floppy disks, hard disks, magneto-optical disks, optical disks (e.g., CD-ROMs, CD-R, CD-RWs, DVD-ROMs, DVD-RAMs, DVD-RWs), magnetic tapes, nonvolatile memory cards, and ROMs. Alternatively, the program code may be downloaded from a server computer or from the cloud via a communications network.
The foregoing description has been directed to specific embodiments of this disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
Not all steps and elements in the above flows and system structure diagrams are necessary, and some steps or elements may be omitted according to actual needs. The execution order of the steps is not fixed, and can be determined as required. The apparatus structures described in the above embodiments may be physical structures or logical structures, that is, some units may be implemented by the same physical entity, or some units may be implemented by a plurality of physical entities, or some units may be implemented by some components in a plurality of independent devices.
The term "exemplary" used throughout this specification means "serving as an example, instance, or illustration," and does not mean "preferred" or "advantageous" over other embodiments. The detailed description includes specific details for the purpose of providing an understanding of the described technology. However, the techniques may be practiced without these specific details. In some instances, well-known structures and devices are shown in block diagram form in order to avoid obscuring the concepts of the described embodiments.
Although the embodiments of the present disclosure have been described in detail with reference to the accompanying drawings, the embodiments of the present disclosure are not limited to the specific details of the embodiments, and various simple modifications may be made to the technical solutions of the embodiments of the present disclosure within the technical spirit of the embodiments of the present disclosure, and all of them fall within the scope of the embodiments of the present disclosure.
The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the scope of the disclosure. Thus, the description is not intended to be limited to the examples and designs described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (16)
1. A method of sample domain conversion of training samples of a recommendation model, comprising:
acquiring a first training sample under a first sample domain and a second training sample under a second sample domain; wherein the training samples in the first sample domain are used for training a first recommendation model for recommending a first class of commodities, the training samples in the second sample domain are used for training a second recommendation model for recommending a second class of commodities, the first training samples comprise samples for the first class of commodities, and the second training samples comprise samples for the second class of commodities;
migrating the second training sample into a third training sample using a first sample migration transformation process; and
determining the third training sample as a training sample under the first sample domain, such that the first recommendation model is trained based on the determined training samples,
wherein the first sample migration transition process is implemented using a machine learning method under a first constraint comprising: and the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a first threshold value, wherein the distributed distance of each migration conversion is the distributed distance between the third training sample obtained by the migration conversion and the first training sample.
2. The method of claim 1, wherein the first training sample and the second training sample are supervised training samples, and
the first constraint further comprises: the training error of the third training sample is less than a second threshold, and the prediction error of the first training sample is less than a third threshold,
wherein the predicted value of the first training sample is predicted using a first prediction model, the first prediction model being trained using the third training sample.
3. The method of claim 1 or 2, further comprising:
using a second sample migration transformation process to transform the third training sample migration into a fourth training sample, wherein the second sample migration transformation process is implemented using a machine learning method under a second constraint; and
determining the third training sample as a training sample under the first sample domain comprises:
determining the third training sample as a training sample under the first sample domain when the fourth training sample satisfies a third constraint,
wherein the third constraint includes: a distributed distance between the fourth training sample and the second training sample is less than a fourth threshold.
4. The method of claim 3, wherein the second training sample and the fourth training sample are supervised training samples, an
The third constraint further comprises: the training error of the fourth training sample is less than a fifth threshold, and the prediction error of the second training sample is less than a sixth threshold,
wherein the predicted value of the second training sample is predicted using a second prediction model, the second prediction model being trained using the fourth training sample.
5. The method of claim 3, further comprising:
and when the fourth training sample does not meet a third constraint condition, determining the fourth training sample as a second training sample under a second sample domain, and returning to execute the operation of converting the second training sample into a third training sample by using the first sample migration conversion process.
6. The method of claim 3, wherein the second constraint comprises:
number of cycles, or
And the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a seventh threshold value, wherein the distributed distance of each migration conversion is the distributed distance between the fourth training sample and the second training sample obtained by the migration conversion.
7. The method of claim 1, wherein the first sample domain is a sample domain corresponding to a few-sample scene and the second sample domain is a sample domain corresponding to a multiple-sample scene.
8. The method of any of claim 6, wherein the distributed distance comprises at least one of:
euclidean distance;
k L divergence;
JS divergence; and
the EM distance.
9. An apparatus for sample domain conversion of training samples of a recommendation model, comprising:
the training sample acquisition unit is used for acquiring a first training sample in a first sample domain and a second training sample in a second sample domain; wherein the training samples in the first sample domain are used for training a first recommendation model for recommending a first class of commodities, the training samples in the second sample domain are used for training a second recommendation model for recommending a second class of commodities, the first training samples comprise samples for the first class of commodities, and the second training samples comprise samples for the second class of commodities;
a first sample migration conversion unit that migration-converts the second training sample into a third training sample using a first sample migration conversion process; and
a first training sample determination unit that determines the third training sample as a training sample in the first sample domain so that the first recommendation model is trained based on the determined training sample,
wherein the first sample migration transition process is implemented using a machine learning method under a first constraint comprising: and the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a first threshold value, wherein the distributed distance of each migration conversion is the distributed distance between the third training sample obtained by the migration conversion and the first training sample.
10. The apparatus of claim 9, wherein the first training sample and the second training sample are supervised training samples, and
the first constraint further comprises: the training error of the third training sample is less than a second threshold, and the prediction error of the first training sample is less than a third threshold,
wherein the predicted value of the first training sample is predicted using a first prediction model, the first prediction model being trained using the third training sample.
11. The apparatus of claim 9 or 10, further comprising:
a second sample migration conversion unit that migrates and converts the third training sample into a fourth training sample using a second sample migration conversion process, wherein the second sample migration conversion process is implemented using a machine learning method under a second constraint condition; and
the first training sample determination unit:
determining the third training sample as a training sample under the first sample domain when the fourth training sample satisfies a third constraint,
wherein the third constraint includes: a distributed distance between the fourth training sample and the second training sample is less than a fourth threshold.
12. The apparatus of claim 11, wherein the second training sample and the fourth training sample are supervised training samples, an
The third constraint further comprises: the training error of the fourth training sample is less than a fifth threshold, and the prediction error of the second training sample is less than a sixth threshold,
wherein the predicted value of the second training sample is predicted using a second prediction model, the second prediction model being trained using the fourth training sample.
13. The apparatus of claim 11, further comprising:
and the second training sample determining unit is used for determining the fourth training sample as a second training sample in a second sample domain when the fourth training sample does not meet a third constraint condition, and triggering the first sample migration and conversion unit to execute the operation of migrating and converting the second training sample into a third training sample by using the first sample migration and conversion process.
14. The apparatus of claim 11, wherein the second constraint comprises:
number of cycles, or
And the difference value between the distributed distance of the current migration conversion and the distributed distance of the last migration conversion is smaller than a seventh threshold value, wherein the distributed distance of each migration conversion is the distributed distance between the fourth training sample and the second training sample obtained by the migration conversion.
15. An electronic device, comprising:
at least one processor, and
a memory coupled with the at least one processor, the memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform the method of any of claims 1-8.
16. A machine-readable storage medium storing executable instructions that, when executed, cause the machine to perform the method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010205866.1A CN111428783B (en) | 2020-03-23 | 2020-03-23 | Method and device for performing sample domain conversion on training samples of recommendation model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010205866.1A CN111428783B (en) | 2020-03-23 | 2020-03-23 | Method and device for performing sample domain conversion on training samples of recommendation model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111428783A true CN111428783A (en) | 2020-07-17 |
| CN111428783B CN111428783B (en) | 2022-06-21 |
Family
ID=71548506
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010205866.1A Active CN111428783B (en) | 2020-03-23 | 2020-03-23 | Method and device for performing sample domain conversion on training samples of recommendation model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111428783B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881355A (en) * | 2020-07-28 | 2020-11-03 | 北京深演智能科技股份有限公司 | Object recommendation method and device, storage medium and processor |
| CN112199603A (en) * | 2020-12-07 | 2021-01-08 | 平安科技（深圳）有限公司 | Information pushing method and device based on countermeasure network and computer equipment |
| CN112396547A (en) * | 2021-01-21 | 2021-02-23 | 平安科技（深圳）有限公司 | Course recommendation method, device, equipment and medium based on unsupervised learning |
| CN113705683A (en) * | 2021-08-30 | 2021-11-26 | 北京达佳互联信息技术有限公司 | Recommendation model training method and device, electronic equipment and storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106548210A (en) * | 2016-10-31 | 2017-03-29 | 腾讯科技（深圳）有限公司 | Machine learning model training method and device |
| CN108564066A (en) * | 2018-04-28 | 2018-09-21 | 国信优易数据有限公司 | A kind of person recognition model training method and character recognition method |
| CN109359793A (en) * | 2018-08-03 | 2019-02-19 | 阿里巴巴集团控股有限公司 | A kind of prediction model training method and device for new scene |
| CN110070076A (en) * | 2019-05-08 | 2019-07-30 | 北京字节跳动网络技术有限公司 | Method and apparatus for choosing trained sample |
| CN110377587A (en) * | 2019-07-15 | 2019-10-25 | 腾讯科技（深圳）有限公司 | Method, apparatus, equipment and medium are determined based on the migrating data of machine learning |
| CN110598845A (en) * | 2019-08-13 | 2019-12-20 | 中国平安人寿保险股份有限公司 | Data processing method, data processing device, computer equipment and storage medium |
| CN110738476A (en) * | 2019-09-24 | 2020-01-31 | 支付宝(杭州)信息技术有限公司 | sample migration method, device and equipment |
-
2020
- 2020-03-23 CN CN202010205866.1A patent/CN111428783B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106548210A (en) * | 2016-10-31 | 2017-03-29 | 腾讯科技（深圳）有限公司 | Machine learning model training method and device |
| US20190318202A1 (en) * | 2016-10-31 | 2019-10-17 | Tencent Technology (Shenzhen) Company Limited | Machine learning model training method and apparatus, server, and storage medium |
| CN108564066A (en) * | 2018-04-28 | 2018-09-21 | 国信优易数据有限公司 | A kind of person recognition model training method and character recognition method |
| CN109359793A (en) * | 2018-08-03 | 2019-02-19 | 阿里巴巴集团控股有限公司 | A kind of prediction model training method and device for new scene |
| CN110070076A (en) * | 2019-05-08 | 2019-07-30 | 北京字节跳动网络技术有限公司 | Method and apparatus for choosing trained sample |
| CN110377587A (en) * | 2019-07-15 | 2019-10-25 | 腾讯科技（深圳）有限公司 | Method, apparatus, equipment and medium are determined based on the migrating data of machine learning |
| CN110598845A (en) * | 2019-08-13 | 2019-12-20 | 中国平安人寿保险股份有限公司 | Data processing method, data processing device, computer equipment and storage medium |
| CN110738476A (en) * | 2019-09-24 | 2020-01-31 | 支付宝(杭州)信息技术有限公司 | sample migration method, device and equipment |
Non-Patent Citations (2)
| Title |
|---|
| TIANYANG WANG 等: "Instance-based Deep Transfer Learning", 《ARXIV:1809.02776V2[CS.CV]》 * |
| 展希希鸿: "机器学习之迁移学习", 《BLOG.CSDN.NET/QQ_28266311/ARTICLE/DETAILS/83057059》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881355A (en) * | 2020-07-28 | 2020-11-03 | 北京深演智能科技股份有限公司 | Object recommendation method and device, storage medium and processor |
| CN112199603A (en) * | 2020-12-07 | 2021-01-08 | 平安科技（深圳）有限公司 | Information pushing method and device based on countermeasure network and computer equipment |
| CN112199603B (en) * | 2020-12-07 | 2021-03-09 | 平安科技（深圳）有限公司 | Information pushing method and device based on countermeasure network and computer equipment |
| CN112396547A (en) * | 2021-01-21 | 2021-02-23 | 平安科技（深圳）有限公司 | Course recommendation method, device, equipment and medium based on unsupervised learning |
| CN113705683A (en) * | 2021-08-30 | 2021-11-26 | 北京达佳互联信息技术有限公司 | Recommendation model training method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111428783B (en) | 2022-06-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111428783B (en) | Method and device for performing sample domain conversion on training samples of recommendation model | |
| Fajtl et al. | Summarizing videos with attention | |
| US11521221B2 (en) | Predictive modeling with entity representations computed from neural network models simultaneously trained on multiple tasks | |
| CN105224623B (en) | The training method and device of data model | |
| Wang et al. | Efficient video transformers with spatial-temporal token selection | |
| CN112381216B (en) | Training and predicting method and device for mixed graph neural network model | |
| KR20150080090A (en) | System for recommending product based on Hadoop and apparatus and methode for recommending product in the system | |
| CN110046784A (en) | A kind of risk of user's access determines method and device | |
| CN112199600A (en) | Target object identification method and device | |
| CN112613938A (en) | Model training method and device and computer equipment | |
| CN113032676B (en) | Recommendation method and system based on micro-feedback | |
| CN112801425B (en) | Method and device for determining information click rate, computer equipment and storage medium | |
| CN108563648B (en) | Data display method and device, storage medium and electronic device | |
| CN113657087A (en) | Information matching method and device | |
| Kang et al. | Autoencoder-based graph construction for semi-supervised learning | |
| CN117034963A (en) | Model training method, translation method, prediction method, recommendation method and device | |
| CN112434746A (en) | Pre-labeling method based on hierarchical transfer learning and related equipment thereof | |
| CN115659055A (en) | Commodity recommendation method, system, equipment and storage medium based on event sequence | |
| CN113240071B (en) | Method and device for processing graph neural network, computer equipment and storage medium | |
| CN115269998A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN114331602A (en) | Model training method based on transfer learning, information recommendation method and device | |
| CN114996578A (en) | Model training method, target object selection method, device and electronic equipment | |
| CN113239215A (en) | Multimedia resource classification method and device, electronic equipment and storage medium | |
| CN115329183A (en) | Data processing method, device, storage medium and equipment | |
| CN116028617B (en) | Information recommendation method, apparatus, device, readable storage medium and program product |
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 |