WO2021082695A1 - 一种训练方法、特征提取方法、装置及电子设备 - Google Patents

一种训练方法、特征提取方法、装置及电子设备 Download PDF

Info

Publication number
WO2021082695A1
WO2021082695A1 PCT/CN2020/111799 CN2020111799W WO2021082695A1 WO 2021082695 A1 WO2021082695 A1 WO 2021082695A1 CN 2020111799 W CN2020111799 W CN 2020111799W WO 2021082695 A1 WO2021082695 A1 WO 2021082695A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature set
neural network
short
target
term
Prior art date
Application number
PCT/CN2020/111799
Other languages
English (en)
French (fr)
Inventor
李怀松
潘健民
Original Assignee
支付宝(杭州)信息技术有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 支付宝(杭州)信息技术有限公司 filed Critical 支付宝(杭州)信息技术有限公司
Publication of WO2021082695A1 publication Critical patent/WO2021082695A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • This document relates to the field of data processing technology, in particular to a training method, feature extraction method, device and electronic equipment.
  • the purpose of the embodiments of this specification is to provide a training method, a feature extraction method, and related devices, which can train a model that can correlate short-term characteristics and long-term characteristics with higher efficiency.
  • a training method including: inputting a first short-term feature set under a target classification corresponding to a sample object into a recurrent neural network to obtain a second short-term feature set, wherein each short-term feature set in the first short-term feature set The features correspond to the same first time granularity; the second short-term feature set is combined into a long-term feature set in chronological order, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity, and the first Second, the time granularity is greater than the first time granularity; input the long-term feature set to a convolutional neural network to obtain the target feature set corresponding to the target object under the target classification; input the target feature set to Recognizing the classification model of the target classification to train the recurrent neural network and the convolutional neural network based on the recognition result of the classification model for the sample object.
  • a feature extraction method including: inputting a first short-term feature set of a target object under a target classification into a recurrent neural network to obtain a second short-term feature set, wherein each of the first short-term feature set The short-term features correspond to the same first time granularity; the second short-term feature set is combined into a long-term feature set in chronological order, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity, and The second time granularity is greater than the first time granularity; the long-term feature set is input to the convolutional neural network to obtain the target feature set under the target classification corresponding to the target object; wherein, the recurrent neural network and the The convolutional neural network is to input the target feature set of the sample object into the classification model that recognizes the target classification, and obtain the recognition result for the sample object based on the classification model, and compare the recurrent neural network and the A convolutional neural network is obtained through training
  • a neural network training device including: a first processing module, which inputs a first short-term feature set under a target classification corresponding to a sample object into a recurrent neural network to obtain a second short-term feature set; wherein, the Each short-term feature in the first short-term feature set corresponds to the same first time granularity; the first combination module combines the second short-term feature set into a long-term feature set in chronological order, wherein each long-term feature set in the long-term feature set The features correspond to the same second time granularity, and the second time granularity is greater than the first time granularity; the second processing module inputs the long-term feature set to the convolutional neural network to obtain that the target object corresponds to the A target feature set under target classification; a training module, which inputs the target feature set to a classification model for identifying the target classification, so as to perform an evaluation on the cyclic nerve based on the recognition result of the classification model for the sample object
  • the network including: a first processing
  • an electronic device including: a memory, a processor, and a computer program stored in the memory and capable of running on the processor, the computer program being executed by the processor: corresponding sample objects
  • the first short-term feature set under the target classification is input to the recurrent neural network to obtain the second short-term feature set; wherein, each short-term feature in the first short-term feature set corresponds to the same first time granularity; and the second short-term feature set corresponds to the same first time granularity;
  • the feature sets are combined into a long-term feature set in chronological order, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity, and the second time granularity is greater than the first time granularity; and the long-term feature
  • the target feature set is input to the convolutional neural network to obtain the target feature set under the target classification corresponding to the target object; the target feature set is input to the classification model used to identify the target classification, so as to target the target object based on
  • a computer-readable storage medium is provided, and a computer program is stored on the computer-readable storage medium.
  • the short-term feature set is input to the recurrent neural network to obtain the second short-term feature set; wherein, each short-term feature in the first short-term feature set corresponds to the same first time granularity; the second short-term feature set is combined in time sequence
  • the growth period feature set wherein each long-term feature in the long-term feature set corresponds to the same second time granularity, and the second time granularity is greater than the first time granularity; and the long-term feature set is input to the convolutional nerve
  • the network obtains the target feature set under the target classification corresponding to the target object; inputting the target feature set to a classification model for identifying the target classification to identify the sample object based on the classification model As a result, the recurrent neural network and the convolutional neural network are trained.
  • a feature extraction device including: a third processing module, which inputs a first short-term feature set of a target object under a target classification to a recurrent neural network to obtain a second short-term feature set; wherein, the first short-term feature set is Each short-term feature in the short-term feature set corresponds to the same first time granularity; the second combination module combines the second short-term feature set into a long-term feature set in chronological order, wherein each long-term feature in the long-term feature set corresponds to Have the same second time granularity, the second time granularity is greater than the first time granularity; a fourth processing module, input the long-term feature set to the convolutional neural network, and obtain the target object corresponding to the target classification The target feature set under the following; wherein, the recurrent neural network and the convolutional neural network are the recognition results obtained based on the classification model after the target feature set of the sample object is input into the classification model that recognizes the target classification
  • an electronic device including: inputting a first short-term feature set of a target object under a target classification into a recurrent neural network to obtain a second short-term feature set; wherein each short-term feature set in the first short-term feature set The features correspond to the same first time granularity; the second short-term feature set is combined into a long-term feature set in chronological order, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity, and the first The second time granularity is greater than the first time granularity; the long-term feature set is input to the convolutional neural network to obtain the target feature set under the target classification corresponding to the target object; wherein, the recurrent neural network and the The convolutional neural network is to input the target feature set of the sample object into the classification model that recognizes the target classification, and then train the recurrent neural network and the convolutional neural network based on the recognition result obtained by the classification model The obtained target feature set of the
  • a computer-readable storage medium is provided, and a computer program is stored on the computer-readable storage medium.
  • the short-term feature set is input to the recurrent neural network to obtain the second short-term feature set; wherein, each short-term feature in the first short-term feature set corresponds to the same first time granularity; the second short-term feature set is combined in time sequence The growth period feature set, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity, and the second time granularity is greater than the first time granularity; and the long-term feature set is input to the convolutional nerve Network to obtain the target feature set under the target classification corresponding to the target object; wherein, the recurrent neural network and the convolutional neural network input the target feature set of the sample object into a classification capable of recognizing the target classification After the model, based on the recognition results obtained by the classification model, the recurrent neural network and
  • the solution in the embodiment of this specification adopts the RNN+CNN model structure.
  • short-term features are formed into long-term features, and the long-term features are further converted into single-dimensional target features and then input to the classifier, so as to be based on the output of the classifier
  • the parameters of RNN and CNN are adjusted to achieve the purpose of training.
  • the whole training process uses both short-term features and long-term features, which not only greatly improves training efficiency, but also enables the model to learn the invisible connection between short-term features and long-term features, thereby obtaining better model performance.
  • FIG. 1 is a schematic flowchart of the training method provided by the embodiment of the specification.
  • Fig. 2 is a schematic diagram of the training structure in the training method provided by the embodiment of the specification.
  • FIG. 3 is a schematic diagram of the steps of the feature extraction method provided by the embodiment of the specification.
  • Fig. 4 is a schematic diagram of the structure of the training device provided by the embodiment of the specification.
  • Fig. 5 is a schematic structural diagram of a feature extraction device provided by an embodiment of this specification.
  • Fig. 6 is a schematic structural diagram of an electronic device provided by an embodiment of the specification.
  • the prior art model training method is to separately train the model (the model is composed of a neural network) according to the characteristics of different time granularities. For example, first input short-term features into the model, and adjust the model parameters according to the output results. After that, the long-term features are further input to the model, and the model parameters are adjusted according to the output results. In this way, firstly, the training efficiency is not high; secondly, although the entire model is based on short-term and long-term features for learning, the training process is completely independent, and the implicit association between short-term and long-term features cannot be formed, resulting in The model cannot achieve better performance after training.
  • this document aims to provide a technical solution that can simultaneously train the model with short-term features and long-term features. Further, it also provides a technical solution for realizing related applications based on the trained model.
  • Fig. 1 is a flowchart of a training method according to an embodiment of this specification. The method shown in FIG. 1 can be executed by the corresponding device below, and includes steps S102 to S108.
  • Step S102 Input the first short-term feature set under the target classification corresponding to the sample object to the Recurrent Neural Network (RNN) to obtain a second short-term feature set.
  • RNN Recurrent Neural Network
  • Each short-term feature set in the first short-term feature set corresponds to the same first short-term feature set.
  • the recurrent neural network is part of the model to be trained.
  • the first short-term features may be relatively intuitive short-term features of the sample object, and these short-term features can be obtained through a relatively conventional feature extraction method, and the embodiment of this specification does not specifically limit the obtaining method.
  • the purpose of inputting the first short-term feature set to the RNN is to refine the first short-term feature set by the RNN to obtain the implicit second short-term feature set.
  • the short-term features in the second short-term feature set may have the same time granularity corresponding to the short-term features in the first short-term feature set, that is, the first time granularity.
  • Step S104 Combine the second short-term feature set into a long-term feature set in a chronological order.
  • Each long-term feature in the long-term feature set corresponds to the same second time granularity, and the second time granularity is greater than the first time granularity.
  • the long-term characteristics are composed of short-term characteristics, so not only the long-term characteristics of the sample object can be extracted, but also the short-term characteristics of the sample object can be extracted.
  • the first time granularity and the second time granularity can be flexibly set according to actual needs, which are not specifically limited in the embodiment of this specification.
  • the second short-term feature set includes daily short-term features of the sample object. This step specifically combines the short-term characteristics of the sample object for 7 days adjacent to each other to obtain the long-term characteristics of the sample object for one week.
  • Step S106 Input the long-term feature set to Convolutional Neural Networks (CNN) to obtain the target feature set under the target classification corresponding to the target object.
  • CNN Convolutional Neural Networks
  • CNN as a part of the model to be trained, has roughly the same purpose as the above-mentioned RNN, and can further refine the long-term feature set to obtain a higher-order target feature set.
  • step S108 the target feature set is input to the classification model for identifying the target classification, so as to train the recurrent neural network and the convolutional neural network based on the recognition result of the classification model for the sample object.
  • the classification model is a part that needs to be cited during training, and is not limited to be a part of the model to be trained.
  • the training method is not unique and depends on the specific structure of the classification model.
  • this step can train the classification model based on a supervised training method. That is, the target feature set is used as the input for identifying the classification model, and the label of the sample object (the label is used to indicate whether the sample object meets the target classification) is used as the output of the classification model to identify the result of the sample object based on the classification model.
  • Train RNN and CNN are supervised training methods.
  • this step can train the classification model based on an unsupervised training method.
  • the unsupervised training method does not need to use labels, so in this step, the target feature set can be directly used as the input for identifying the classification model, so as to train the RNN and CNN based on the recognition result of the classification model for the sample object.
  • the classification model can also be trained based on the recognition result, thereby improving the recognition accuracy of the classification model and ensuring the training effect of RNN and CNN.
  • the scheme of the embodiment of this specification adopts the RNN+CNN model structure.
  • short-term features are formed into long-term features, and the long-term features are further converted into single-dimensional target features.
  • the whole training process uses both short-term features and long-term features, which not only greatly improves training efficiency, but also enables the model to learn the invisible connection between short-term features and long-term features, thereby obtaining better model performance.
  • the training method of the embodiment of this specification uses both short-term features and long-term features to train the target model.
  • the training structure includes: RNN ⁇ CNN ⁇ classification model.
  • RNN+CNN belongs to the target model to be trained, and the classification model is a temporary part added during the training process and is not part of the target model.
  • the training method of the embodiment of this specification first inputs the first short-term feature set under the target classification corresponding to the sample object to the RNN, and obtains the second short-term feature set output by the RNN.
  • the RNN described here may be any one of a long-term and short-term memory network, a gated recurrent unit network, and a self-attention mechanism network, or may include: a long-term short-term memory network, a gated recurrent unit network, and a self-attention mechanism network At least one of them. Since RNN belongs to the prior art, this article will not go into details.
  • the RNN does not change the time granularity of the short-term features, so the short-term features in the second short-term feature set obtained by input may have the same time granularity corresponding to the short-term features in the first short-term feature set.
  • the short-term features in the second short-term feature set can be combined in chronological order to obtain corresponding long-term features with greater time granularity.
  • a vector combination method can be used to combine short-term features into long-term features. For example: Combine short-term features A (q, w, e) and short-term features B (a, s, d) to obtain long-term features AB (q, w, e, a, s, d). It should be understood that long-term features are spliced from short-term features, and therefore contain short-term features of the sample object.
  • the combined long-term features are input to CNN, and the target feature set is further refined by CNN.
  • CNN has different implementation modes, which are not specifically limited in the embodiment of this specification.
  • a CNN may include: a convolutional layer, a pooling layer, and a fully connected layer.
  • the convolution layer is used to perform convolution processing on the long-term feature set to obtain the output feature set of the convolution layer.
  • the pooling layer is used to pool the output feature set of the convolutional layer based on the maximum pooling algorithm and/or the mean pooling algorithm to obtain the output feature set of the pooling layer.
  • the fully connected layer is used to convert the output feature set of the pooling layer into a single-dimensional target feature set suitable for the classification model.
  • the target feature of the target feature set can be input to the classification model, and the sample object is classified by the classification model to identify whether the sample object meets the target classification.
  • the recognition result output by the classification model belongs to the training result, and the training result is not necessarily the real result.
  • calculate the loss between the training result and the real result according to the loss function and adjust the parameters of RNN, CNN and the classification model for the purpose of reducing the loss (it is not necessary to adjust the parameters of the classifier, depending on the classification model Is there any need for adjustment) in order to achieve the purpose of training.
  • This application scenario is used to train a learning model that characterizes financial risks.
  • the learning model adopts the structure of Long Short-Term Memory (LSTM) + Text-CNN (Text-CNN), and the corresponding process includes the following steps.
  • LSTM Long Short-Term Memory
  • Text-CNN Text-CNN
  • Step 1 Obtain the financial business data of the sample objects in the payment application, and based on the semantic analysis algorithm, extract the basic features of the financial business data at each half-small time granularity to obtain the first short-term feature set of a month.
  • the first short-term feature set can be, but is not limited to, the total transaction amount, the total number of transactions, and the total number of counterparties corresponding to the sample object every half hour. These characterize the transaction behavior of the sample objects in a short period of time. Some abnormal transaction patterns (such as fast forward and fast exit) can be captured by these short-term characteristics.
  • Step 2 Input the first short-term feature set to the LSTM, and obtain the second short-term feature set output by the LSTM.
  • the number of LSTMs is not limited to one.
  • the LSTM may have a one-to-one correspondence with the days of the first short-term feature set, so that the output of each Lstm represents a day's short-term hidden features.
  • Step 3 Combine the second short-term feature set in chronological order to obtain a long-term feature set.
  • the short-term hidden features of every half hour were previously obtained, but they can only represent the trading dynamics of half an hour.
  • the short-term hidden features of half an hour are spliced into the long-term daily in chronological order. feature. It should be understood that the long-term feature data format should be applicable to subsequent TextCnn.
  • Step 4 Input the long-term feature set into TextCnn, and then extract the target feature set from TextCnn.
  • the length of the convolutional layer of TextCnn can be set freely. For example, if the length is 2, it can capture the local behavior changes of the sample object in the adjacent 2 days. If the length is 6, the local behavior changes of the adjacent 6 days can be captured. That is to say, through the combination of different sizes of the convolution kernel, the feature learning of the sample objects at different time granularities is realized.
  • the pooling layer of TextCnn performs a Pooling operation on the output features extracted by the convolution.
  • the pooling layer can use the Max-Pooling algorithm and the Avg-Pooling algorithm at the same time.
  • Max-Pooling is mainly used to retain the main information of feature changes
  • Avg-Pooling is used to retain the average state of features.
  • the fully connected layer of TextCnn integrates the feature set obtained by the Pooling operation to reduce the dimensionality, and obtains a single-dimensional target feature set suitable for the input classification model.
  • Step 5 Input the target feature set into the classification model to train LSTM and TextCnn.
  • the classification model can adopt a two-class cross-entropy mechanism.
  • the loss function of the classification model is specifically the cross-entropy loss function, and the label value of the sample object can only be 1 or 0. 1 indicates that the sample object meets the target classification, and 0 indicates that the sample object does not meet the target classification.
  • the loss function can be: -log(yt
  • yp) -[yt*log(yp)+(1-yt) log(1-yp)]. For the entire learning model, its loss function is the non-negative average of the loss functions of all sample objects.
  • the classification model After the target feature set is input into the classification model, the classification model will obtain the recognition result of whether the sample object belongs to the risk object. After that, the loss of the recognition result and the label value is calculated based on the loss function, and the parameters of LSTM and TextCnn are adjusted for the purpose of reducing the loss.
  • an embodiment of this specification also provides a feature extraction method, including steps 302 to 306.
  • Step 302 Input the first short-term feature set of the target object under the target classification to the recurrent neural network to obtain a second short-term feature set.
  • Each short-term feature in the first short-term feature set corresponds to the same first time granularity.
  • Step 304 Combine the second short-term feature set into a long-term feature set in chronological order, and each long-term feature in the long-term feature set corresponds to the same second time granularity, and the second time granularity is greater than the first time granularity.
  • Step 306 Input the long-term feature set to a convolutional neural network to obtain a target feature set under the target classification corresponding to the target object.
  • the target features in the target feature set are the hidden features of the target object finally refined.
  • the above-mentioned cyclic neural network and the above-mentioned convolutional neural network are trained by the training method shown in FIG. 1. That is, the recurrent neural network and the convolutional neural network input the target feature set of the sample object into the classification model that recognizes the target classification, and obtain the recognition result for the sample object based on the classification model, It is obtained by training the recurrent neural network and the convolutional neural network, and the target feature set of the sample object is determined based on the recurrent neural network and the convolutional neural network.
  • the solution of the embodiment of this specification only needs to input the short-term features of the target object into the RNN+CNN model, that is, it is extracted from the model mechanically and presents both short-term characteristics and long-term characteristics.
  • the target characteristics of can be used to describe the target object more comprehensively, and dig out hidden characteristics that are difficult to find manually.
  • an embodiment of this specification also provides a neural network training device 400, which includes the following modules.
  • the first processing module 410 inputs the first short-term feature set under the target classification corresponding to the sample object to the recurrent neural network to obtain the second short-term feature set; wherein, each short-term feature in the first short-term feature set corresponds to the same first short-term feature set.
  • One time granularity is one time granularity.
  • the first combination module 420 combines the second short-term feature set into a long-term feature set in chronological order, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity, and the second time granularity is greater than The first time granularity.
  • the second processing module 430 inputs the long-term feature set to a convolutional neural network to obtain a target feature set under the target classification corresponding to the target object.
  • the training module 440 inputs the target feature set to a classification model for recognizing the target classification, so as to compare the recurrent neural network and the convolutional neural network based on the recognition result of the classification model for the sample object.
  • the network is trained.
  • the solution of the embodiment of this specification adopts the RNN+CNN model structure.
  • short-term features are formed into long-term features, and the long-term features are further converted into single-dimensional target features.
  • the whole training process uses both short-term features and long-term features, which not only greatly improves training efficiency, but also enables the model to learn the invisible connection between short-term features and long-term features, thereby obtaining better model performance.
  • the target feature set is specifically used as the input of the classification model for identifying the target classification
  • the label of the sample object is used as the output of the classification model to be based on the
  • the classification model trains the recurrent neural network and the convolutional neural network according to the recognition result of the sample object, wherein the label of the sample object is used to indicate whether the sample object meets the target classification .
  • the recurrent neural network includes at least one of the following: a long and short-term memory network, a gated recurrent unit network, and a self-attention mechanism network.
  • the convolutional neural network includes: a text rolling recurrent network.
  • the convolutional neural network includes: a convolutional layer, which performs convolution processing on a long-term feature set to obtain an output feature set of the convolutional layer; a pooling layer, based on a maximum pooling algorithm and/or an average pooling algorithm , Performing pooling processing on the output feature set of the convolutional layer to obtain the output feature set of the pooling layer; the fully connected layer converts the output feature set of the pooling layer into a single-dimensional target feature set.
  • the sample object is a payment application user
  • the target classification is a financial risk
  • the first short-term feature set includes short-term features of at least one of the following feature dimensions: the payment application user has a first time granularity Corresponding total transaction amount, total number of transactions, and total number of counterparties.
  • the training device of the embodiment of the present specification can be used as the execution body of the training method shown in FIG. 1 above, and therefore can realize the functions realized by the training method in FIG. 1 and FIG. 2. Since the principle is the same, this article will not repeat them.
  • an embodiment of this specification also provides a feature extraction device, which includes the following modules.
  • the third processing module 510 inputs the first short-term feature set under the target classification to the recurrent neural network to obtain a second short-term feature set; wherein, each short-term feature in the first short-term feature set corresponds to the same first short-term feature set.
  • One time granularity is one time granularity.
  • the second combination module 520 combines the second short-term feature set into a long-term feature set in chronological order, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity, and the second time granularity is greater than The first time granularity.
  • the fourth processing module 530 inputs the long-term feature set to a convolutional neural network to obtain a target feature set under the target classification corresponding to the target object.
  • the recurrent neural network and the convolutional neural network input the target feature set of the sample object into a classification model that recognizes the target classification, and then, based on the recognition result obtained by the classification model, analyze the recurrent neural network.
  • the network and the convolutional neural network are trained, and the target feature set of the sample object is determined based on the recurrent neural network and the convolutional neural network.
  • the solution of the embodiment of this specification only needs to input the short-term features of the target object into the RNN+CNN model, that is, it is extracted by the model mechanically and presents both short-term characteristics and long-term characteristics.
  • the target characteristics of can be used to describe the target object more comprehensively, and dig out hidden characteristics that are difficult to find manually.
  • the feature extraction device of the embodiment of the present specification can be used as the execution subject of the feature extraction method shown in FIG. 3, and therefore can realize the function of the feature extraction method in FIG. 3. Since the principle is the same, this article will not repeat them.
  • FIG. 6 is a schematic diagram of the structure of an electronic device according to an embodiment of this specification.
  • the electronic device includes a processor, and optionally an internal bus, a network interface, and a memory.
  • the memory may include memory, such as high-speed random access memory (Random-Access Memory, RAM), or may also include non-volatile memory (non-volatile memory), such as at least one disk storage.
  • RAM Random-Access Memory
  • non-volatile memory such as at least one disk storage.
  • the electronic device may also include hardware required by other services.
  • the processor, network interface, and memory can be connected to each other through an internal bus.
  • the internal bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect, peripheral component interconnection standard) bus, or an EISA (Extended) bus. Industry Standard Architecture, extended industry standard structure) bus, etc.
  • the bus can be divided into an address bus, a data bus, a control bus, and so on. For ease of presentation, only one bidirectional arrow is used to indicate in FIG. 6, but it does not mean that there is only one bus or one type of bus.
  • the program may include program code, and the program code includes computer operation instructions.
  • the memory may include memory and non-volatile memory, and provide instructions and data to the processor.
  • the processor reads the corresponding computer program from the non-volatile memory to the memory and then runs it to form a neural network training device on the logical level.
  • the processor executes the program stored in the memory, and is specifically configured to perform the following operations: input the first short-term feature set under the target classification corresponding to the sample object to the recurrent neural network to obtain the second short-term feature set, wherein the first short-term feature set Each short-term feature in the short-term feature set corresponds to the same first time granularity; the second short-term feature set is combined into a long-term feature set in chronological order, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity.
  • the second time granularity is greater than the first time granularity
  • the long-term feature set is input to the convolutional neural network to obtain the target feature set under the target classification corresponding to the target object
  • the target The feature set is input to a classification model for recognizing the target classification, so as to train the recurrent neural network and the convolutional neural network based on the recognition result of the classification model for the sample object.
  • the processor reads the corresponding computer program from the non-volatile memory to the memory and then runs it, which can also form a feature extraction device at a logical level.
  • the processor executes the program stored in the memory, and is specifically configured to perform the following operations: input the first short-term feature set of the target object under the target classification to the recurrent neural network to obtain the second short-term feature set, wherein the first short-term feature set is Each short-term feature in the short-term feature set corresponds to the same first time granularity; the second short-term feature set is combined into a long-term feature set in chronological order, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity.
  • the second time granularity is greater than the first time granularity;
  • the long-term feature set is input to the convolutional neural network to obtain the target feature set under the target classification corresponding to the target object;
  • the The cyclic neural network and the convolutional neural network input the target feature set of the sample object into a classification model that recognizes the target classification, and then obtain the recognition result for the sample object based on the classification model, and perform A neural network and the convolutional neural network are trained, and the target feature set of the sample object is determined based on the cyclic neural network and the convolutional neural network.
  • the above-mentioned training method disclosed in the embodiment shown in FIG. 1 or the feature extraction method disclosed in the embodiment shown in FIG. 3 is implemented by a processor.
  • the processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or instructions in the form of software.
  • the above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (Network Processor, NP), etc.; it may also be a digital signal processor (DSP), a dedicated integrated Circuit (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.
  • CPU central processing unit
  • NP Network Processor
  • DSP digital signal processor
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • the methods, steps, and logic block diagrams disclosed in the embodiments of this specification can be implemented or executed.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.
  • the steps of the method disclosed in the embodiments of this specification can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of hardware and software modules in the decoding processor.
  • the software module can be located in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, registers.
  • the storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.
  • the electronic device of the embodiment of this specification can realize the function of the above-mentioned training apparatus in the embodiment shown in FIG. 1 and FIG. 2 or the function of the above-mentioned feature extraction apparatus in the embodiment shown in the figure. Since the principle is the same, this article will not repeat them.
  • the electronic equipment in this specification does not exclude other implementations, such as logic devices or a combination of software and hardware, etc. That is to say, the execution body of the following processing flow is not limited to each logic unit. It can also be a hardware or logic device.
  • the embodiment of this specification also proposes a computer-readable storage medium that stores one or more programs, and the one or more programs include instructions.
  • the portable electronic device can execute the training method of the embodiment shown in FIG. 1, and is specifically used to execute the following method: classify the sample object corresponding to the target
  • the first short-term feature set of is input to the recurrent neural network to obtain the second short-term feature set, wherein each short-term feature in the first short-term feature set corresponds to the same first time granularity;
  • the second short-term feature set is Time sequence is combined into a long-term feature set, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity, and the second time granularity is greater than the first time granularity;
  • the long-term feature set is input to
  • the convolutional neural network obtains the target feature set under the target classification corresponding to the target object;
  • the target feature set is input to the classification model for identifying the target classification, so as to target the sample based on the classification model Training the recurrent neural network and the convolutional neural network for the recognition result
  • the portable electronic device can execute the feature extraction method of the embodiment shown in FIG. 3, and is specifically used to execute the following method: classify the target object under the target classification
  • the first short-term feature set of is input to the recurrent neural network to obtain the second short-term feature set, wherein each short-term feature in the first short-term feature set corresponds to the same first time granularity;
  • the second short-term feature set is Time sequence is combined into a long-term feature set, wherein each long-term feature in the long-term feature set corresponds to the same second time granularity, and the second time granularity is greater than the first time granularity;
  • the long-term feature set is input to
  • the convolutional neural network obtains the target feature set under the target classification corresponding to the target object; wherein, the recurrent neural network and the convolutional neural network input the target feature set of the sample object to the target feature set to recognize the target After the classified classification model, the recognition result for the sample object
  • this specification can be provided as a method, a system or a computer program product. Therefore, this specification may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
  • computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

本说明书实施例提供一种训练方法、特征提取方法、装置及电子设备。训练方法包括：将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集，其中，第一短期特征集中的各短期特征对应有相同的第一时间粒度。将第二短期特征集按照时间顺序组合成长期特征集，其中，长期特征集中的各长期特征对应有相同的第二时间粒度，第二时间粒度大于第一时间粒度。将长期特征集输入至卷积神经网络，得到目标对象对应所述目标分类下的目标特征集。将目标特征集输入至用于识别目标分类的分类模型，以基于所述分类模型针对样本对象的识别结果，对循环神经网络和卷积神经网络进行训练。

Description

一种训练方法、特征提取方法、装置及电子设备 技术领域
本文件涉及数据处理技术领域，尤其涉及一种训练方法、特征提取方法、装置及电子设备。
背景技术
随着人工智能的发展，越来越多的场景会应用到由神经网络所构建的深度学习模型，以达到机械化处理信息的目的。在其中一些场景中，需要使用不同时间粒度所呈现的特征对模型进行训练。现有技术的作为是分别针对每种时间粒度的特征，对模型进行单独训练。这种方式下，首先训练效率不高；其次，训练后的模型无法体现出短期特性与长期特性之间的隐性关联，导致模型性能不佳。
有鉴于此，如何以较高的效率，训练出能够关联短期特性和长期特性的模型，是当前亟需要解决的技术问题。
发明内容
本说明书实施例目的是提供一种训练方法、特征提取方法及相关装置，能够以较高的效率，训练出能够关联短期特性和长期特性的模型。
为了实现上述目的，本说明书实施例是通过以下方面实现的。
第一方面，提供一种训练方法，包括：将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集，其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
第二方面，提供一种特征提取方法，包括：将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集，其中，所述第一短期特征集中的各 短期特征对应有相同的第一时间粒度；将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；其中，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
第三方面，提供一种神经网络的训练装置，包括：第一处理模块，将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；第一组合模块，将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；第二处理模块，将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；训练模块，将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
第四方面，提供一种电子设备包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述计算机程序被所述处理器执行：将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
第五方面，提供一种算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现如下步骤：将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；将所述第二短期特征集按照时间顺序组合 成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
第六方面，提供一种特征提取装置，包括：第三处理模块，将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；第二组合模块，将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；第四处理模块，将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；其中，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
第七方面，提供一种电子设备，包括：将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；其中，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
第八方面，提供一种算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现如下步骤：将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；将所述长期特征集输入至卷积神经网络，得到 所述目标对象对应所述目标分类下的目标特征集；其中，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
本说明书实施例的方案采用RNN+CNN的模型结构，在训练过程中，将短期特征组成长期特征，并进一步将长期特征转换为单维度的目标特征后输入至分类器，从而根据分类器的输出结果调整RNN和CNN的参数，以达到训练目的。显然，整个训练过程同时使用了短期特征和长期特征，不仅大幅提高了训练效率，还能够使模型学习到短期特征和长期特征之间的隐形联系，从而获得更好的模型性能。
附图说明
为了更清楚地说明本说明书实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本说明书实施例中记载的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本说明书实施例提供的训练方法的流程示意图。
图2为本说明书实施例提供的训练方法中的训练结构示意图。
图3为本说明书实施例提供的特征提取方法的步骤示意图。
图4为本说明书实施例提供的训练装置的结构示意图。
图5为本说明书实施例提供的特征提取装置的结构示意图。
图6为本说明书实施例提供的电子设备的结构示意图。
具体实施方式
为了使本技术领域的人员更好地理解本说明书中的技术方案，下面将结合本说明书实施例中的附图，对本说明书实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本说明书一部分实施例，而不是全部的实施例。基于本说明书中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都应当属于本说明书保护的范围。
如前所述，现有技术的模型训练方法是针对不同时间粒度的特征，单独对模型(模型由神经网络组成)进行训练。比如，先将短期特征输入至模型，并根据输出结果对模型参数进行调整。之后，再进一步将长期特征输入至模型，并根据输出结果对模型参数进行调整。这种方式下，首先训练效率不高；其次，整个模型虽然是基于短期特征和长期特征进行了学习，但是训练过程是完全独立的，无法形成短期特征和长期特征之间的隐性关联，导致模型训练后达不到较佳的性能。
针对上述问题，本文件旨在提供一种可以将短期特征和长期特征同时对模型进行训练的技术方案。进一步地，还提供基于训练后的模型实现相关应用的技术方案。
图1是本说明书实施例训练方法的流程图。图1所示的方法可以由下文相对应的装置执行，包括步骤S102至S108。
步骤S102，将样本对象对应目标分类下的第一短期特征集输入至循环神经网络(RNN，Recurrent Neural Network)，得到第二短期特征集，第一短期特征集中的各短期特征对应有相同的第一时间粒度。
其中，循环神经网络作为待训练模型中的一部分。第一短期特征可以是比较直观的样本对象的短期特征，这些短期特征可以通过较为常规的特征提取方式获取得到，本说明书实施例不对获取方法作具体限定。
本步骤中，将第一短期特征集输入至RNN的目的是由RNN对第一短期特征集进行提炼，得到隐性的第二短期特征集。第二短期特征集中的短期特征可以与第一短期特征集中的短期特征对应有相同的时间粒度，即第一时间粒度。
步骤S104，将第二短期特征集按照时间顺序组合成长期特征集，长期特征集中的各长期特征对应有相同的第二时间粒度，第二时间粒度大于第一时间粒度。
显然，长期特征是通过短期特征组合而成的，因此不仅可以提现出样本对象的长期特性，也能够提现出样本对象的短期特性。
此外，应理解的是，第一时间粒度和第二时间粒度可以根据实际需要进行灵活设置，本说明书实施例不作具体限定。作为示例性介绍，假设第一时间粒度为一天、第二时间粒度为一周，则第二短期特征集中包含有样本对象每天的短期特征。本步骤具体将样本对象相邻7天的短期特征进行组合，得到样本对象一周的长期特征。
步骤S106，将长期特征集输入至卷积神经网络(CNN，Convolutional Neural Networks)，得到目标对象对应目标分类下的目标特征集。
其中，CNN作为待训练模型中的一部分，与上述RNN的用途大致相同，可对长期特征集作进一步提炼，获得更高阶的目标特征集。
步骤S108，将目标特征集输入至用于识别目标分类的分类模型，以基于分类模型针对样本对象的识别结果，对循环神经网络和卷积神经网络进行训练。
其中，分类模型是训练时所需要引用的部分，并不限定作为待训练模型的一部分。
此外，训练方式并不唯一，取决于分类模型的具体结构。
如果分类模型采用的是分类器结构，则本步骤可以基于有监督的训练方式对分类模型进行训练。即，将目标特征集作为用于识别分类模型的输入，将样本对象的标签(标签用于指示样本用对象是否符合目标分类)作为分类模型的输出，以基于分类模型针对样本对象的识别结果，对RNN和CNN进行训练。
如果分类模型采用的是解码器结构，则本步骤可以基于无监督的训练方式对分类模型进行训练。无监督的训练方式不需要使用标签，因此本步骤可以直接将目标特征集作为用于识别分类模型的输入，以基于分类模型针对样本对象的识别结果，对RNN和CNN进行训练。此外，在训练过程中，还可以基于识别结果，对分类模型进行训练，从而提高分类模型的识别准确率，保证RNN和CNN的训练效果。
基于图1所示的训练方法可以知道，本说明书实施例的方案采用RNN+CNN的模型结构，在训练过程中，将短期特征组成长期特征，并进一步将长期特征转换为单维度的目标特征后输入至分类器，从而根据分类器的输出结果调整RNN和CNN的参数，以达到训练目的。显然，整个训练过程同时使用了短期特征和长期特征，不仅大幅提高了训练效率，还能够使模型学习到短期特征和长期特征之间的隐形联系，从而获得更好的模型性能。
下面对说明书实施例的训练方法进行详细介绍。
本说明书实施例的训练方法同时使用短期特征和长期特征对目标模型进行训练。如图2所示，训练结构包括：RNN→CNN→分类模型。其中，RNN+CNN属于待训练的目标模型，分类模型是训练过程中添加的临时部分，并不作为目标模型的一部分。
本说明书实施例的训练方法首先将样本对象对应目标分类下的第一短期特征集输入至RNN，得到由RNN输出的第二短期特征集。
这里所述的RNN可以是长短期记忆网络、门控循环单元网络以及自注意力机制网 络中的任一者，或者，可以包括：长短期记忆网络、门控循环单元网络以及自注意力机制网络中的至少一者。由于RNN属于现有技术，本文不再具体赘述。
应理解，RNN并不会改变短期特征的时间粒度，因此输入获得的第二短期特征集中的短期特征可以与第一短期特征集中的短期特征对应有相同的时间粒度。
在获得RNN输出的第二短期特征集后，即可按照时间顺序对第二短期特征集中的短期特征进行组合，得到对应有更大时间粒度的长期特征。
这里需要说明的是，特征的组合方法并不唯一，本说明书实施例不作具体限定。作为其中一种可行的方案，可以采用向量组合方式将短期特征组合成长期特征。比如：将短期特征A(q,w,e)和短期特征B(a,s,d)进行组合，可以得到的长期特征AB(q,w,e,a,s,d)。应理解，长期特征是由短期特征拼接而成的，因此含有样本对象短期的特性。
之后，将组合而成的长期特征输入至CNN，由CNN进一步提炼出的目标特征集。
应理解，CNN与RNN一样，具有不同的实现方式，本说明书实施例不作具体限定。
作为示例性介绍，CNN可以包括：卷积层、池化层和全连接层。卷积层用于对长期特征集进行卷积处理，得到卷积层输出特征集。池化层用于基于最大值池化算法和/或均值池化算法，对卷积层输出特征集进行池化处理，得到池化层输出特征集。全连接层用于将池化层输出特征集转换为单一维度的适用于分类模型的目标特征集。
在获得目标特征集后，即可将目标特征集的目标特征输入至分类模型，由分类模型对样本对象进行分类，以识别样本对象是否符合目标分类。
这里，样本对象是否符合目标分类属于已知信息，分类模型输出的识别结果属于训练结果，训练结果并不一定是真实结果。之后，根据损失函数来计算训练结果与真实结果之间的损失，并以降低损失为目的，对RNN、CNN以及分类模型的参数进行调整(也可以不对分类器的参数进行调整，取决于分类模型是否有调整需求)，以达到训练目的。
下面结合一个实际的应用场景，对本说明书实施例的训练方法进行实例介绍。
本应用场景用于训练刻画金融风险特征的学习模型。其中，学习模型采用长短期记忆网络(LSTM，Long Short-Term Memory)+文本卷进循环网络(Text-CNN)的结构，对应的流程包括以下步骤。
步骤一，获取支付应用中样本对象的金融业务数据，并基于语义分析算法，按照每半小的时间粒度，对金融业务数据进行基础特征的提取，得到一个月的第一短期特征集。
在本应用场景中，第一短期特征集可以但不限于是样本对象每半小时所对应的交易总金额、交易总笔数以及交易对手总数。这些刻画的是样本对象在短时间内的交易行为，一些不正常的交易模式(如快进快出)可以被这些短期特征捕捉到。
步骤二，将第一短期特征集输入至LSTM，得到LSTM输出的第二短期特征集。
其中，LSTM数量并不限于一个。作为示例性介绍，LSTM可以与第一短期特征集的天数一一对应，这样每个Lstm的输出代表了一天的短期隐藏特征。
步骤三，将第二短期特征集按照时间顺序进行组合，得到长期特征集。
如前所述，之前获取了每半小时的短期隐藏特征，但是只能代表半小时的交易动态，为了得到样本对象长期的交易动态，按时间顺序将半小时的短期隐藏特征拼接成每天的长期特征。应理解，长期特征的数据格式应适用于后续的TextCnn。
步骤四，将长期特征集输入至TextCnn，由TextCnn提炼出目标特征集。
其中，TextCnn的卷积层长度可以自由设置，比如长度为2则可以捕获样本对象相邻2天的局部行为变化，如果长度为6，可以捕捉相邻6天的局部行为变化。也就是说，通过卷积核不同尺寸的组合实现对样本对象不同时间粒度的特征学习。
TextCnn的池化层对卷积提的输出特征再进行Pooling操作。本应用场景中，池化层可以同时采用最大值池化(Max-Pooling)算法与(Avg-Pooling)算法。其中，Max-Pooling主要用来保留特征发生变化的主要信息，Avg-Pooling用来保留特征平均状态。
TextCnn的全连接层将Pooling操作得到的特征集进行整合降维，得到适合输入分类模型的单一维度的目标特征集。
步骤五，将目标特征集输入至分类模型，以对LSTM和TextCnn进行训练。
其中，分类模型可以采用二分类交叉熵机制。在二分类问题中，分类模型的损失函数具体为交叉熵损失函数，样本对象的标签取值只能是1或0,1表示样本对象符合目标分类，0表示样本对象不符合目标分类。
假设某个样本对象的真实标签为yt,该样本对象yt＝1的概率为yp,则损失函数可以为：-log(yt|yp)＝-[yt*log(yp)+(1-yt)log(1-yp)]。对于整个学习模型而言，其损失函数就是所有样本对象的损失函数非负的平均值。
目标特征集输入在输入分类模型后，会得到分类模型识别样本对象是否属于风险对象的识别结果。之后，基于损失函数计算识别结果会与标签取值的损失，并以降低损失 为目的，来调整LSTM和TextCnn的参数。
以上是对本说明书实施例的方法的介绍。应理解，在不脱离本文上述原理基础之上，还可以进行适当的变化，这些变化也应视为本说明书实施例的保护范围。
此外，如图3所示，本说明书实施例还提供一种特征提取方法，包括步骤302至306。
步骤302，将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集，第一短期特征集中的各短期特征对应有相同的第一时间粒度。
步骤304，将第二短期特征集按照时间顺序组合成长期特征集，长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度。
步骤306，将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集。
其中，目标特征集中的目标特征即最终提炼得到的目标对象的隐性特征。
应理解，上述循环神经网络和上述卷积神经网络是由图1所示的训练方法所训练得到的。即，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
基于图3所示的特征提取方法可以知道，本说明书实施例的方案仅需要将目标对象的短期特征输入至RNN+CNN的模型，即由模型机械方式提炼出即呈现短期特性，又呈现长期特性的目标特征，可用于对目标对象进行更全面的刻画，挖掘出人工难以找到的隐性特征。
此外，如图4所示，本说明书实施例还提供一种神经网络的训练装置400，包括以下模块。
第一处理模块410，将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度。
第一组合模块420，将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度。
第二处理模块430，将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集。
训练模块440，将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
基于图4所示的训练装置可以知道，本说明书实施例的方案采用RNN+CNN的模型结构，在训练过程中，将短期特征组成长期特征，并进一步将长期特征转换为单维度的目标特征后输入至分类器，从而根据分类器的输出结果调整RNN和CNN的参数，以达到训练目的。显然，整个训练过程同时使用了短期特征和长期特征，不仅大幅提高了训练效率，还能够使模型学习到短期特征和长期特征之间的隐形联系，从而获得更好的模型性能。
可选地，训练模块440在执行时，具体将所述目标特征集作为用于识别所述目标分类的分类模型的输入，将所述样本对象的标签作为所述分类模型的输出，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练，其中，所述样本对象的标签用于指示所述样本用对象是否符合所述目标分类。
可选地，所述循环神经网络包括以下至少一者：长短期记忆网络、门控循环单元网络以及自注意力机制网络。
可选地，所述卷积神经网络包括：文本卷进循环网络。
可选地，所述卷积神经网络包括：卷积层，对长期特征集进行卷积处理，得到卷积层输出特征集；池化层，基于最大值池化算法和/或均值池化算法，对所述卷积层输出特征集进行池化处理，得到池化层输出特征集；全连接层，将池化层输出特征集转换为单一维度的目标特征集。
可选地，所述样本对象为支付应用用户，所述目标分类为金融风险，所述第一短期特征集包括以下至少一种特征维度的短期特征：所述支付应用用户在各第一时间粒度所对应的交易总金额、交易总笔数以及交易对手总数。
显然，本说明书实施例的训练装置可以作为上述图1所示的训练方法的执行主体，因此能够实现该训练方法在图1和图2所实现的功能。由于原理相同，本文不再赘述。
此外，如图5所示，本说明书实施例还提供一种特征提取装置，包括以下模块。
第三处理模块510，将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度。
第二组合模块520，将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度。
第四处理模块530，将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集。
其中，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
基于图5所示的特征提取装置可以知道，本说明书实施例的方案仅需要将目标对象的短期特征输入至RNN+CNN的模型，即由模型机械方式提炼出即呈现短期特性，又呈现长期特性的目标特征，可用于对目标对象进行更全面的刻画，挖掘出人工难以找到的隐性特征。
显然，本说明书实施例的特征提取装置可以作为上述图3所示的特征提取方法的执行主体，因此能够实现该特征提取方法在图3所实现的功能。由于原理相同，本文不再赘述。
图6是本说明书的一个实施例电子设备的结构示意图。请参考图6，在硬件层面，该电子设备包括处理器，可选地还包括内部总线、网络接口、存储器。其中，存储器可能包含内存，例如高速随机存取存储器(Random-Access Memory，RAM)，也可能还包括非易失性存储器(non-volatile memory)，例如至少1个磁盘存储器等。当然，该电子设备还可能包括其他业务所需要的硬件。
处理器、网络接口和存储器可以通过内部总线相互连接，该内部总线可以是ISA(Industry Standard Architecture，工业标准体系结构)总线、PCI(Peripheral Component Interconnect，外设部件互连标准)总线或EISA(Extended Industry Standard Architecture，扩展工业标准结构)总线等。所述总线可以分为地址总线、数据总线、控制总线等。为便于表示，图6中仅用一个双向箭头表示，但并不表示仅有一根总线或一种类型的总线。
存储器，用于存放程序。具体地，程序可以包括程序代码，所述程序代码包括计算机操作指令。存储器可以包括内存和非易失性存储器，并向处理器提供指令和数据。
其中，处理器从非易失性存储器中读取对应的计算机程序到内存中然后运行，在逻辑层面上形成神经网络的训练装置。处理器，执行存储器所存放的程序，并具体用于执行以下操作：将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集，其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
其中，处理器从非易失性存储器中读取对应的计算机程序到内存中然后运行，在逻辑层面上还可以形成特征提取装置。处理器，执行存储器所存放的程序，并具体用于执行以下操作：将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集，其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；其中，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
上述如本说明书图1所示实施例揭示的训练方法或者图3所示实施例揭示的特征提取方法由处理器实现。处理器可能是一种集成电路芯片，具有信号的处理能力。在实现过程中，上述方法的各步骤可以通过处理器中的硬件的集成逻辑电路或者软件形式的指令完成。上述的处理器可以是通用处理器，包括中央处理器(Central Processing Unit，CPU)、网络处理器(Network Processor，NP)等；还可以是数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field－Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。可以实现或者执行本说明书实施例中的公 开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。结合本说明书实施例所公开的方法的步骤可以直接体现为硬件译码处理器执行完成，或者用译码处理器中的硬件及软件模块组合执行完成。软件模块可以位于随机存储器，闪存、只读存储器，可编程只读存储器或者电可擦写可编程存储器、寄存器等本领域成熟的存储介质中。该存储介质位于存储器，处理器读取存储器中的信息，结合其硬件完成上述方法的步骤。
应理解，本说明书实施例的电子设备可以实现上述训练装置在图1和图2所示的实施例的功能，或者上述特征提取装置在图所示的实施例的功能。由于原理相同，本文不再赘述。
当然，除了软件实现方式之外，本说明书的电子设备并不排除其他实现方式，比如逻辑器件抑或软硬件结合的方式等等，也就是说以下处理流程的执行主体并不限定于各个逻辑单元，也可以是硬件或逻辑器件。
此外，本说明书实施例还提出了一种计算机可读存储介质，该计算机可读存储介质存储一个或多个程序，该一个或多个程序包括指令。
其中，该指令当被包括多个应用程序的便携式电子设备执行时，能够使该便携式电子设备执行图1所示实施例的训练方法，并具体用于执行以下方法：将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集，其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
或者，指令当被包括多个应用程序的便携式电子设备执行时，能够使该便携式电子设备执行图3所示实施例的特征提取方法，并具体用于执行以下方法：将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集，其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；其中，所述循环 神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
应理解，上述指令当被包括多个应用程序的便携式电子设备执行时，能够使上文所述的训练装置实现图1和图2所示实施例的功能，或者，能够使上文所述的特征提取装置实现图3所示实施例的功能，本文不再赘述。
本领域技术人员应明白，本说明书的实施例可提供为方法、系统或计算机程序产品。因此，本说明书可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本说明书可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
上述对本说明书特定实施例进行了描述。其它实施例在所附权利要求书的范围内。在一些情况下，在权利要求书中记载的动作或步骤可以按照不同于实施例中的顺序来执行并且仍然可以实现期望的结果。另外，在附图中描绘的过程不一定要求示出的特定顺序或者连续顺序才能实现期望的结果。在某些实施方式中，多任务处理和并行处理也是可以的或者可能是有利的。
以上仅为本说明书的实施例而已，并不用于限制本说明书。对于本领域技术人员来说，本说明书可以有各种更改和变化。凡在本说明书的精神和原理之内所作的任何修改、等同替换、改进等，均应包含在本说明书的权利要求范围之内。此外，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都应当属于本文件的保护范围。

Claims (13)

  1. 一种训练方法，包括：
    将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集，其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；
    将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；
    将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；
    将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
  2. 根据权利要求1所述的方法，
    将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练，包括：
    将所述目标特征集作为用于识别所述目标分类的分类模型的输入，将所述样本对象的标签作为所述分类模型的输出，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练，其中，所述样本对象的标签用于指示所述样本用对象是否符合所述目标分类。
  3. 根据权利要求1所述的方法，还包括：
    所述循环神经网络包括以下至少一者：
    长短期记忆网络、门控循环单元网络以及自注意力机制网络。
  4. 根据权利要求1所述的方法，
    所述卷积神经网络包括：文本卷进循环网络。
  5. 根据权利要求1-4中任一项所述的方法，
    所述卷积神经网络包括：
    卷积层，对长期特征集进行卷积处理，得到卷积层输出特征集；
    池化层，基于最大值池化算法和/或均值池化算法，对所述卷积层输出特征集进行池化处理，得到池化层输出特征集；
    全连接层，将池化层输出特征集转换为单一维度的目标特征集。
  6. 根据权利要求1-4中任一项所述的方法，
    所述目标分类为金融风险，所述第一短期特征集包括以下至少一种特征维度的短期 特征：
    所述样本对象在各第一时间粒度所对应的交易总金额、交易总笔数以及交易对手总数。
  7. 一种特征提取方法，包括：
    将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集，其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；
    将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；
    将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；
    其中，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
  8. 一种神经网络的训练装置，包括：
    第一处理模块，将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；
    第一组合模块，将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；
    第二处理模块，将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；
    训练模块，将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
  9. 一种电子设备包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述计算机程序被所述处理器执行：
    将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；
    将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中 的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；
    将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；
    将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
  10. 一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现如下步骤：
    将样本对象对应目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；
    将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；
    将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；
    将所述目标特征集输入至用于识别所述目标分类的分类模型，以基于所述分类模型针对所述样本对象的识别结果，对所述循环神经网络和所述卷积神经网络进行训练。
  11. 一种特征提取装置，包括：
    第三处理模块，将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；
    第二组合模块，将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；
    第四处理模块，将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；
    其中，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
  12. 一种电子设备包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述计算机程序被所述处理器执行：
    将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期 特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；
    将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；
    将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；
    其中，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
  13. 一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现如下步骤：
    将目标对象属于目标分类下的第一短期特征集输入至循环神经网络，得到第二短期特征集；其中，所述第一短期特征集中的各短期特征对应有相同的第一时间粒度；
    将所述第二短期特征集按照时间顺序组合成长期特征集，其中，所述长期特征集中的各长期特征对应有相同的第二时间粒度，所述第二时间粒度大于所述第一时间粒度；
    将所述长期特征集输入至卷积神经网络，得到所述目标对象对应所述目标分类下的目标特征集；
    其中，所述循环神经网络和所述卷积神经网络是将样本对象的目标特征集输入至具有识别所述目标分类的分类模型后，基于所述分类模型得到的识别结果，对所述循环神经网络和所述卷积神经网络进行训练所得到的，所述样本对象的目标特征集是基于所述循环神经网络和所述卷积神经网络确定得到的。
PCT/CN2020/111799 2019-10-31 2020-08-27 一种训练方法、特征提取方法、装置及电子设备 WO2021082695A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911050651.0 2019-10-31
CN201911050651.0A CN110796240A (zh) 2019-10-31 2019-10-31 一种训练方法、特征提取方法、装置及电子设备

Publications (1)

Publication Number Publication Date
WO2021082695A1 true WO2021082695A1 (zh) 2021-05-06

Family

ID=69442284

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/111799 WO2021082695A1 (zh) 2019-10-31 2020-08-27 一种训练方法、特征提取方法、装置及电子设备

Country Status (3)

Country Link
CN (1) CN110796240A (zh)
TW (1) TWI752486B (zh)
WO (1) WO2021082695A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110796240A (zh) * 2019-10-31 2020-02-14 支付宝(杭州)信息技术有限公司 一种训练方法、特征提取方法、装置及电子设备
CN111899092B (zh) * 2020-08-28 2022-05-20 上海冰鉴信息科技有限公司 基于二道模型的业务数据筛选方法及装置
CN111931690A (zh) * 2020-08-28 2020-11-13 Oppo广东移动通信有限公司 模型训练方法、装置、设备及存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030101050A1 (en) * 2001-11-29 2003-05-29 Microsoft Corporation Real-time speech and music classifier
CN108446978A (zh) * 2018-02-12 2018-08-24 阿里巴巴集团控股有限公司 处理交易数据的方法及装置
CN109451522A (zh) * 2018-09-21 2019-03-08 邵阳学院 一种面向蓝牙网关的流量预测方法及装置
CN110796240A (zh) * 2019-10-31 2020-02-14 支付宝(杭州)信息技术有限公司 一种训练方法、特征提取方法、装置及电子设备

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108416663A (zh) * 2018-01-18 2018-08-17 阿里巴巴集团控股有限公司 评估金融违约风险的方法及装置
CN110084603B (zh) * 2018-01-26 2020-06-16 阿里巴巴集团控股有限公司 训练欺诈交易检测模型的方法、检测方法以及对应装置
CN108734338A (zh) * 2018-04-24 2018-11-02 阿里巴巴集团控股有限公司 基于lstm模型的信用风险预测方法及装置
CN110009384A (zh) * 2019-01-07 2019-07-12 阿里巴巴集团控股有限公司 预测业务指标的方法及装置
CN110084610B (zh) * 2019-04-23 2023-06-23 东华大学 一种基于孪生神经网络的网络交易欺诈检测系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030101050A1 (en) * 2001-11-29 2003-05-29 Microsoft Corporation Real-time speech and music classifier
CN108446978A (zh) * 2018-02-12 2018-08-24 阿里巴巴集团控股有限公司 处理交易数据的方法及装置
CN109451522A (zh) * 2018-09-21 2019-03-08 邵阳学院 一种面向蓝牙网关的流量预测方法及装置
CN110796240A (zh) * 2019-10-31 2020-02-14 支付宝(杭州)信息技术有限公司 一种训练方法、特征提取方法、装置及电子设备

Also Published As

Publication number Publication date
TWI752486B (zh) 2022-01-11
TW202119297A (zh) 2021-05-16
CN110796240A (zh) 2020-02-14

Similar Documents

Publication Publication Date Title
WO2021082695A1 (zh) 一种训练方法、特征提取方法、装置及电子设备
CN107480575B (zh) 模型的训练方法、跨年龄人脸识别方法和对应的装置
CN112395979B (zh) 基于图像的健康状态识别方法、装置、设备及存储介质
CN102156885B (zh) 基于级联式码本生成的图像分类方法
US20210390370A1 (en) Data processing method and apparatus, storage medium and electronic device
CN111653275B (zh) 基于lstm-ctc尾部卷积的语音识别模型的构建方法及装置、语音识别方法
JP2022141931A (ja) 生体検出モデルのトレーニング方法及び装置、生体検出の方法及び装置、電子機器、記憶媒体、並びにコンピュータプログラム
CN110457680A (zh) 实体消歧方法、装置、计算机设备和存储介质
CN111428557A (zh) 基于神经网络模型的手写签名的自动校验的方法和装置
CN112801146A (zh) 一种目标检测方法及系统
CN109948735A (zh) 一种多标签分类方法、系统、装置及存储介质
Yingxin et al. A robust hand gesture recognition method via convolutional neural network
US20210174104A1 (en) Finger vein comparison method, computer equipment, and storage medium
CN110288079A (zh) 特征数据获取方法、装置和设备
CN106910512A (zh) 语音文件的分析方法、装置及系统
CN111695335A (zh) 一种智能面试方法、装置及终端设备
CN112966625A (zh) 一种基于改进的vgg-16网络模型的表情识别方法及装置
CN112232070A (zh) 自然语言处理模型构建方法、系统、电子设备及存储介质
CN111563161A (zh) 一种语句识别方法、语句识别装置及智能设备
CN114925693B (zh) 一种基于多模型融合的多元关系抽取方法和抽取系统
CN113486178B (zh) 文本识别模型训练方法、文本识别方法、装置以及介质
CN115687934A (zh) 意图识别方法、装置、计算机设备及存储介质
CN114639152A (zh) 基于人脸识别的多模态语音交互方法、装置、设备及介质
Zheng et al. Learning from the web: Webly supervised meta-learning for masked face recognition
CN115954019A (zh) 一种融合自注意力和卷积操作的环境噪声识别方法及系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20883654

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20883654

Country of ref document: EP

Kind code of ref document: A1