CN115169551A - Training method of behavior prediction model, risk behavior prediction method and device - Google Patents
Training method of behavior prediction model, risk behavior prediction method and device Download PDFInfo
- Publication number
- CN115169551A CN115169551A CN202210757609.8A CN202210757609A CN115169551A CN 115169551 A CN115169551 A CN 115169551A CN 202210757609 A CN202210757609 A CN 202210757609A CN 115169551 A CN115169551 A CN 115169551A
- Authority
- CN
- China
- Prior art keywords
- behavior
- event
- behavior event
- sample
- behavioral
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Operations Research (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the specification describes a training method of a behavior prediction model, a risk behavior prediction method and a risk behavior prediction device. According to the method of the embodiment, the type identification of the sample behavior event and the time information of the sample behavior event can be obtained when the behavior prediction model is trained. And then, characterizing the sample behavior event in a continuous time domain, and further characterizing and training a behavior prediction model according to the behavior event characterized in the continuous time domain so as to optimize the type identifier output by the behavior prediction model and the predicted value of time. By representing the sample behavior events in a continuous time domain, the correlation between the behavior events and the occurring time of the behavior events is realized, so that the model can fully learn the characteristics of regularity and periodicity presented by the behavior events and the occurring time of the behavior events, and the accuracy of risk behavior prediction can be improved.
Description
Technical Field
One or more embodiments of the present specification relate to the field of artificial intelligence, and in particular, to a training method for a behavior prediction model, a risk behavior prediction method, and an apparatus.
Background
And the data wind control is mainly used for judging the risk of the current behavior event based on the data of the historical behavior event. The method compares the information of the current behavior event with the information of the historical behavior event recorded in the wind control engine, and judges the risk of the current behavior event according to the abnormal degree, so that the method is the most widely used risk prevention and control means in the industry at present.
However, the behavior events of the user usually exhibit corresponding regular or periodic variations in a continuous time domain, which results in a less accurate behavior prediction method based on data at discrete time points.
Disclosure of Invention
One or more embodiments of the present specification describe a training method of a behavior prediction model, a risk behavior prediction method, and an apparatus, which can improve accuracy of risk behavior prediction.
According to a first aspect, there is provided a method of training a behavioral prediction model, comprising:
acquiring type identification of a sample behavior event and time information corresponding to the sample behavior event from historical behavior data; each type of sample behavior event corresponds to a type identifier;
according to the type identification and the time information of the sample behavior event, characterizing the sample behavior event on a continuous time domain to obtain a behavior event characterization;
and training the behavior prediction model according to the behavior event representation so as to optimize the type identification and the predicted value of time output by the behavior prediction model.
In a possible implementation manner, the characterizing the sample behavior event in a continuous time domain according to the type identifier and the time information of the sample behavior event to obtain a behavior event characterization includes:
for each sample behavior event, performing:
mapping the type identifier of the current sample behavior event to a preset space to obtain a mapping vector of the current sample behavior event;
acquiring the behavior event representation of the previous sample behavior event; when the current sample behavior event is a first sample behavior event, the behavior event of the last sample behavior event is characterized as 0;
and determining the behavior event representation of the current sample behavior event according to the mapping vector of the current sample behavior event, the time information corresponding to the current sample behavior event and the behavior event representation of the last sample behavior event.
In a possible implementation manner, the determining, according to the mapping vector of the current sample behavior event, the time information corresponding to the current sample behavior event, and the behavior event characterization of the previous sample behavior event, the behavior event characterization of the current sample behavior event includes:
determining the behavior event characterization of the jth sample behavior event by using the following calculation formula:
h j ＝max{W y *y j +W t *t j +W h *h j-1 +b h ,0}
wherein h is j Behavioral event characterization, y, for characterizing the jth sample behavioral event j Mapping vector, t, for characterizing the j-th sample behavior event j Is used for characterizing the time information corresponding to the j sample behavior event, h j-1 Behavioral event characterization for characterizing the j-1 st sample behavioral event, W y Coefficient, W, for characterizing a linear transformation of behavioral events t For characterizing coefficients that transform time information linearly, W h Coefficients for characterizing the linear transformation of the characterization of the last sample behavior event, b h The correction amount used to characterize behavioral events, max { \8230;, 0} is used to characterize the nonlinear activation function, reLU.
In one possible implementation, the behavior prediction model includes a first prediction probability distribution that predicts a type of a behavior event;
the training of the behavioral prediction model from the behavioral event characterization includes:
determining a first prediction probability distribution for predicting the type of the behavior event according to the behavior event representation;
optimizing model parameters in the first prediction probability distribution using a cross entropy loss function.
In one possible implementation, the determining a first prediction probability distribution for predicting the type of the behavioral event according to the behavioral event characterization includes:
calculating the first prediction probability distribution for predicting the type of the behavior event by using the following calculation formula:
wherein P (k) is used to characterize the first predictive probability distribution, V k A parameter matrix with a type mark k corresponding to a neural network layer for representing the behavior prediction model, h j The behavioral event representation is used for representing the jth sample behavioral event, K is used for representing the type number of the sample behavioral event, b 1 And b 2 All are corrections for adjusting the predicted values.
In one possible implementation, the behavioral prediction model includes a second prediction probability distribution that predicts a time of occurrence of a behavioral event;
the training of the behavioral prediction model from the behavioral event characterization includes:
determining a second prediction probability distribution for predicting the time of occurrence of the behavioral event according to the behavioral event characterization;
model parameters in the second predicted probability distribution are optimized using a logarithmic function.
In a possible implementation manner, the historical behavior data is t i Data prior to the time of day;
determining, from the behavioral event characterization, a second prediction probability distribution that predicts a time of occurrence of a behavioral event, comprising:
determining t from the behavioral event characterization i After the moment, there is a behavior eventA raw first probability density function;
determining from t according to said first probability density function i A second probability density function without any behavior event occurring within a preset time period from the moment;
determining the second predictive probability distribution based on the first probability density function and the second probability density function.
In one possible implementation, the determining t according to the behavioral event characterization i A first probability density function of behavioral events occurring after the time of day, comprising:
determining the first probability density function using the following calculation:
wherein λ is * (t) characterizing said first probability density function, v t Model parameters for linear transformation of behavioral event characterizations, h j Behavioral event characterization, t, for characterizing the jth sample behavioral event j For characterizing the moment of occurrence of the j-th sample behaviour event, b t Correction quantity, w, for characterizing a first probability density function 1 And w 2 All the model parameters are model parameters for transforming time quantity, and p is used for representing the hyper-parameters of the behavior prediction model;
and/or the presence of a gas in the gas,
the determining from t according to the first probability density function i A second probability density function for no behavior events occurring within a predetermined period of time from the time of day, comprising:
calculating the second probability density function using the following calculation:
wherein S is * (t) for characterizing said second probability density function, λ * (t) characterizing the first probability densityA degree function;
and/or the presence of a gas in the atmosphere,
said determining said second predictive probability distribution from said first probability density function and said second probability density function, comprising:
calculating the second predicted probability distribution using the following calculation:
f * (t)＝λ * (t)·S * (t)
wherein, f * (t) for characterizing said second predictive probability distribution, λ * (t) for characterizing the first probability density function, S * (t) is used to characterize the second probability density function.
In one possible implementation, the sample behavior event includes: transaction actions generated by the user; the transaction actions include: at least one of shopping, transportation, dining, and entertainment.
According to a second aspect, there is provided a risk behavior prediction method, comprising:
get the T th p The type identification of at least one behavior event before the moment and the time value corresponding to each behavior event;
inputting the acquired type identifier of the at least one behavior event and the time value corresponding to each behavior event into a behavior prediction model to obtain the Tth behavior event p The type identification predicted value of a behavior event to be predicted after the moment and the time predicted value corresponding to the behavior event to be predicted; wherein the behavior prediction model is obtained by training with the training method of the behavior prediction model according to any one of the first aspect;
searching whether a data set which is consistent with the type identification predicted value and the time predicted value exists in a historical behavior database; at least one data set is stored in the historical behavior database, and each data set comprises a type identifier of a behavior event and a time value corresponding to the behavior event;
and if the behavior event to be predicted is not found, the behavior event to be predicted is a behavior event with risk.
According to a third aspect, there is provided a training apparatus for a behavior prediction model, comprising: the system comprises a sample data acquisition module, a characterization module and a training module;
the first acquisition module is configured to acquire the type identifier of the sample behavior event and the time information corresponding to the sample behavior event from historical behavior data; each type of sample behavior event corresponds to a type identifier;
the characterization module is configured to characterize the sample behavior event in a continuous time domain according to the type identifier and the time information of the sample behavior event acquired by the first acquisition module to obtain a behavior event characterization;
and the training module is configured to train the behavior prediction model according to the behavior event representation obtained by the representation module so as to optimize the type identifier output by the behavior prediction model and the predicted value of time.
According to a fourth aspect, there is provided a risk behavior prediction apparatus comprising: the device comprises a second acquisition module, an input module, a search module and a determination module;
the second obtaining module is configured to obtain the Tth p The type identification of at least one behavior event before the moment and the time value corresponding to each behavior event;
the input module is configured to input the type identifier of the at least one behavior event acquired by the second acquisition module and the time value corresponding to each behavior event into the behavior prediction model to obtain a tth behavior p The type identification predicted value of a behavior event to be predicted after the moment and the time predicted value corresponding to the behavior event to be predicted; wherein the behavior prediction model is trained by using a training device of the behavior prediction model in the third aspect;
the searching module is configured to search whether a data set which is consistent with the type identification predicted value obtained by the input module and is consistent with the time predicted value exists in a historical behavior database; at least one data set is stored in the historical behavior database, and each data set comprises a type identifier of a behavior event and a time value corresponding to the behavior event;
the determining module is configured to determine that the behavior event to be predicted is a risky behavior event if the searching module does not search the behavior event.
According to a fifth aspect, there is provided a computing device comprising: a memory having executable code stored therein, and a processor that, when executing the executable code, implements the method of any of the first and second aspects described above.
According to the method and the device provided by the embodiment of the specification, when the behavior prediction model is trained, the type identification of the sample behavior event and the time information corresponding to the sample behavior event are firstly obtained. The sample behavior events may then be characterized over a continuous time domain according to the type identification and the time information of the sample behavior events. And finally, training a behavior prediction model according to the obtained behavior event representation so as to optimize the type identification and the predicted value of time output by the behavior prediction model. Therefore, the sample behavior events are characterized in a continuous time domain when the model is trained, and the behavior events and the occurring time correlation are realized. Therefore, the behavior prediction model trained based on the obtained behavior event representation can fully learn the rules and periodic changes presented by the behavior event and the occurrence time of the behavior event, so that the accuracy of risk behavior prediction can be improved.
Drawings
In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present specification, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a method for training a behavior prediction model provided in an embodiment of the present description;
FIG. 2 is a sequence diagram of a transaction provided by one embodiment of the present description;
FIG. 3 is a flow diagram of a method for determining behavioral event characterization according to one embodiment of the present description;
FIG. 4 is a flow diagram of a method for training a behavior prediction model according to one embodiment of the present description;
FIG. 5 is a flow diagram of another method for training a behavior prediction model provided in one embodiment of the present description;
FIG. 6 is a flow chart of a method of determining a second predictive probability distribution provided by one embodiment of the present description;
FIG. 7 is a flow diagram of a method for risk behavior prediction according to an embodiment of the present disclosure;
FIG. 8 is a schematic diagram of a training apparatus for a behavior prediction model according to an embodiment of the present disclosure;
fig. 9 is a schematic diagram of a risk behavior prediction apparatus according to an embodiment of the present disclosure.
Detailed Description
As described above, the core idea of prevention and control based on the historical behavior event of the user is to compare the information of the behavior event of this time with the information of the historical behavior event recorded in the wind control engine, and then determine the risk according to the abnormal degree of the event.
In a conventional risk prevention and control scheme, a time interval length is generally preset, a time sequence is converted into a plurality of sections of equal-length time intervals, events in each section of time interval are aggregated, and risk prediction is performed by using obtained characteristics. However, a user's behavioral events will typically exhibit corresponding regular and periodic variations over a continuous time domain. For example, users are 07:00-09:00 are purchased for breakfast at merchants such as convenience stores, dining halls, etc., the user is about to travel between 17 and 30-18. The traditional scheme does not well reserve the time information and the dynamic change model, and loses the relevance between the time information and the event information, so that the prediction accuracy is low.
In the scheme, the behavior event and the time of occurrence of the behavior event are considered to be correlated, the sample behavior event is represented in a continuous time domain, and the rule and the periodic change of the behavior event and the time of occurrence of the behavior event can be fully considered based on the trained behavior prediction model, so that the accuracy of risk behavior prediction can be improved.
As shown in fig. 1, an embodiment of the present specification provides a method for training a behavior prediction model, which may include the following steps:
step 101: acquiring type identification of a sample behavior event and time information corresponding to the sample behavior event from historical behavior data; each type of sample behavior event corresponds to a type identifier;
step 103: according to the type identification and the time information of the sample behavior event, characterizing the sample behavior event on a continuous time domain to obtain behavior event characterization;
step 105: and training a behavior prediction model according to the behavior event characterization so as to optimize the type identification output by the behavior prediction model and the predicted value of time.
When training the behavior prediction model, firstly, the type identification of the sample behavior event and the time information corresponding to the sample behavior event are obtained. The sample behavior events may then be characterized over a continuous time domain according to the type identification and the time information of the sample behavior events. And finally, training a behavior prediction model according to the obtained behavior event representation so as to optimize the type identification and the predicted value of time output by the behavior prediction model. Therefore, the sample behavior events are characterized on the continuous time domain when the model is trained, and the correlation between the behavior events and the occurring time of the behavior events is realized. Therefore, the behavior prediction model trained based on the obtained behavior event representation can fully learn the rules and periodic changes presented by the behavior event and the occurrence time of the behavior event, so that the accuracy of risk behavior prediction can be improved.
The steps in FIG. 1 are described below with reference to specific examples.
First, in step 101, a type identifier of a sample behavior event and time information corresponding to the sample behavior event are obtained from historical behavior data.
The historical behavior data may be data that records the daily behavior of the user. For example, the sample behavior events in the historical behavior data may be user-generated transactions that may include shopping, transportation, dining, entertainment, and the like. Fig. 2 is a sequence diagram of the user's transaction behavior generated at various times, which may include travel, breakfast, lunch, afternoon tea, and Taobao shopping.
Each type of sample behavior event may correspond to a type identifier, for example, a transaction behavior generated by a user traveling in the morning corresponds to a type identifier a, a transaction behavior generated by the user buying breakfast corresponds to a type identifier B, a transaction behavior generated by the user buying things on a shopping APP corresponds to a type identifier C, and the like.
The type identifier corresponding to each type of sample behavior event may encode each sample behavior event according to the type of the sample behavior event to obtain the type identifier of each sample behavior event. For example, the code of "travel" is "0001", "breakfast" is "0010", "lunch" is "0011", "shopping mall" is "0100", "online shopping" is "0101", "sports" is "0110", and so on.
Then, in step 103, according to the type identifier and the time information of the sample behavior event, the sample behavior event is characterized in a continuous time domain to obtain a behavior event characterization.
After the type identification and the corresponding time of each sample behavior event are obtained, the behavior events of each sample are considered to be represented on a continuous time domain. For example, as shown in fig. 3, step 103 may perform the following operations for each sample behavior event:
step 301: mapping the type identifier of the current sample behavior event to a preset space to obtain a mapping vector of the current sample behavior event;
step 303: acquiring the behavior event representation of the behavior event of the previous sample; when the current sample behavior event is a first sample behavior event, the behavior event of the last sample behavior event is characterized as 0;
step 305: and determining the behavior event representation of the current sample behavior event according to the mapping vector of the current sample behavior event, the time information corresponding to the current sample behavior event and the behavior event representation of the last sample behavior event.
In this embodiment, when each sample behavior event is characterized in a continuous time domain, the type identifier of the current sample behavior event may be mapped to a preset space. And then acquiring the behavior event representation of the previous sample behavior event, and determining the behavior event representation of the current sample behavior event according to the obtained mapping vector of the current sample behavior event, the time information corresponding to the current sample behavior event and the behavior event representation of the previous sample behavior event. Therefore, the sample behavior events are correlated with the time information, the influence of the previous sample behavior event on the current sample behavior event is integrated into the sample behavior events, and the representation of each sample behavior event on the time domain is fully considered, so that the prediction accuracy can be improved when the possible future behavior events of the user are predicted.
Step 301 will be explained below.
In this step, after each sample behavior event is coded to obtain a type identifier, the type identifier is considered to be mapped into an embedded vector, so that parameters such as a dimensional coefficient of mapping can be adjusted according to the size of a vector to be processed, and each type identifier is mapped into a preset spatial dimension. For example, in one possible implementation, the type identifier may be mapped using the following calculation:
y j ＝W em *z j +b em
wherein, y j Mapping vector, W, for characterizing the j-th sample behavior event em For the mapping coefficients, z can be determined according to the size of the spatial dimension to be mapped j Is the jthType identification of sample behavioral events, b em Is the mapped correction amount.
Step 303 is explained below.
The behavior event representation of the jth sample behavior event in the cycle layer needs to consider the behavior event representation of the jth-1 sample behavior event, so the behavior event representation of the last sample behavior event needs to be acquired. Of course, if the current sample behavior event is the first sample behavior event of the input loop layer training behavior prediction model, the behavior event of the last sample behavior event is characterized as 0. That is, when the current sample behavior event is determined, the current sample behavior event is determined according to the mapping vector of the current sample behavior event and the time information corresponding to the current sample behavior event.
Step 305 is explained below.
After the mapping vector of the current sample behavior event, the time information of the current sample behavior event and the behavior event characterization of the previous sample behavior event are obtained, the behavior event characterization of the current sample behavior event is determined by simultaneously utilizing linear change and a nonlinear activation function. For example, in one possible implementation, step 305 may determine the behavior event characterization of the jth sample behavior event using the following calculation:
h j ＝max{W y *y j +W t *t j +W h *h j-1 +b h ,0}
wherein h is j Behavioral event characterization, y, for characterizing the jth sample behavioral event j Mapping vector for characterizing the jth sample behavior event, t j Is used for representing the time information corresponding to the j sample behavior event, h j-1 Behavioral event characterization for characterizing the j-1 st sample behavioral event, W y Coefficient for characterizing a linear transformation of a behavioral event, W t For characterizing coefficients that transform time information linearly, W h Coefficients for characterizing the linear transformation of the characterization of the last sample behavior event, b h The correction amount used to characterize behavioral events, max { \8230;, 0} is used to characterize the nonlinear activation function, reLU.
From the above formula, W y 、W t And W h The method is used for linear transformation, and fully fuses the type identification, the event information and the characterization of the last sample behavior event. And max { \8230;, 0} is characterized by a nonlinear activation function ReLU, namely, a current behavior event is characterized by utilizing linear and nonlinear transformation simultaneously, so that the obtained behavior event characterization can more comprehensively cover the characteristic information of the current sample behavior event. Note that the ellipses in max { \8230, 0} are the ones before the commas in the brackets, for example, in the above calculation formula for behavior event characterization, the ellipses represent W y *y j +W t *t j +W h *h j-1 +b h . And if the current sample behavior event is the first sample behavior event in the loop layer of the input neural network, the ellipsis is W y *y j +W t *t j +b h 。
Further in step 105, the behavior prediction model is trained according to the behavior event characterization to optimize the type identifier and the predicted value of time output by the behavior prediction model.
In embodiments of the present description, the behavior prediction model may output predictions for type identification, and may also output predictions for time. Aiming at the type identification, the behavior prediction model gives prediction according to a first prediction probability distribution for predicting the type of the behavior event; for time, the behavior prediction model gives a prediction according to a second prediction probability distribution for predicting the time of occurrence of the behavior event, and the two cases are described below.
The first condition is as follows: the behavior prediction model comprises a first prediction probability distribution for predicting the type of the behavior event;
as shown in fig. 4, in the present case, step 105 may be implemented by the following steps when training the behavior prediction model:
step 401: determining a first prediction probability distribution for predicting the type of the behavior event according to the behavior event representation;
step 403: model parameters in the first prediction probability distribution are optimized using a cross entropy loss function.
In this embodiment, when a behavior prediction model is trained according to a behavior event representation, a first prediction probability distribution for predicting the type of a behavior event is determined according to the behavior event representation, and then model parameters in the first prediction probability distribution are optimized by using a cross entropy loss function. Therefore, the first prediction probability distribution of the type of the predicted behavior event is continuously optimized, so that more optimal model parameters are obtained, and the behavior prediction model is more reliable in predicting the type of the behavior event.
Step 401 in determining the first predictive probability distribution may be determined using the following calculation:
wherein P (k) is used to characterize a first predictive probability distribution, V k A parameter matrix with a type mark k corresponding to a neural network layer for representing a behavior prediction model, h j The behavioral event representation is used for representing the jth sample behavioral event, K is used for representing the type number of the sample behavioral event, b 1 And b 2 All are corrections for adjusting the predicted values.
Step 403 aims at minimizing the cross-entropy loss function when optimizing the model parameters in the first prediction probability distribution using the cross-entropy loss function. Specifically, on the basis of the cross entropy loss function, in each iteration process, the values of the loss function are used for back propagation, and model parameters in the first prediction probability distribution are updated until an iteration stop condition is reached. Wherein the iteration stop condition may be, for example, a loss function convergence, a preset number threshold reached by the number of iterations, etc.
Case two: the behavioral prediction model includes a second predictive probability distribution that predicts a time of occurrence of the behavioral event.
As shown in fig. 5, in the present case, the step 105 may be implemented by the following steps when training the behavior prediction model:
step 501: determining a second prediction probability distribution for predicting the time of occurrence of the behavioral event according to the behavioral event representation;
step 503: model parameters in the second predicted probability distribution are optimized using a logarithmic function.
In this embodiment, when the behavior prediction model is trained according to the behavior event characterization, a second prediction probability distribution for predicting the occurrence time of the behavior event is determined according to the behavior event characterization, and then model parameters in the second prediction probability distribution are optimized by using a logarithmic function. In this way, by continuously optimizing the second prediction probability distribution for predicting the occurrence time of the behavior event, a better model parameter is obtained, so that the reliability of the behavior prediction model is higher when the occurrence time of the behavior event is predicted.
In this embodiment, since each behavior event is characterized in a continuous time domain, a modeling means based on a time sequence point process is considered to determine the second prediction probability distribution. The second predictive probability distribution describes the probability density that no behavioral event has occurred for a certain period of time and that a behavioral event has occurred at time t in the future. Then, a second predictive probability distribution may be further determined based on the time-point process by determining a probability density function of no behavior event occurring within a certain time period and a probability density function of behavior event occurring after a certain time, respectively.
For example, as shown in FIG. 6, when the historical behavior data is t i In the case of data prior to the time of day, step 501 may determine the second predictive probability distribution by:
step 601: determining t from behavioral event characterization i A first probability density function of occurrence of a behavioral event after the time instant;
step 603: from the first probability density function, determine t i A second probability density function without any behavior event occurring in a preset time period from the moment;
step 605: a second predictive probability distribution is determined based on the first probability density function and the second probability density function.
In this embodiment, when determining the second prediction probability distribution for predicting the occurrence time of the behavior event, first, t may be determined according to the behavior event characterization i There is a first probability density function of the occurrence of a behavioral event after the time of day. Then, a self t is determined according to the first probability density function i And a second probability density function of no action event occurring within a preset time period from the moment. Thus, the joint probability density of the first probability density function and the second probability density function represents t i And no action event occurs within t, and the probability density function of the action event occurring at the future time t is the second prediction probability distribution.
Step 601 will be explained below.
The core of the time sequence point process is the conditional strength function lambda thereof * (t), the first probability density function. If an infinitesimally small time window t, t + dt is given]Based on the historical events before time t, there are: h t ＝{t i ,z i |t i T is less than or equal to t }, wherein z is i Represents the time t i The type of behavioral event that occurred identifies. Then the probability of a future occurrence of a behavioral event can be formally defined as: lambda [ alpha ] * (t)dt＝P(eventin[t,t+dt]|H t )＝E(dN(t)|H t ) Wherein E (dN (t) | H t ) Representation based on historical events H t In the time window [ t, t + dt]The expectation of the number of events occurring within, N (t) represents the number of events occurring before time t. If it is assumed that two events do not occur at the same time, i.e., dN (t) e {0,1}, then omitting a given condition yields λ * (t), the main difference between the different point process models is λ * (t) difference in form.
In the present application, consider determining the first probability density function using the following equation:
wherein λ is * (t) for characterizing the first probability density function,v t model parameters for linear transformation of behavioral event characterization, h j Behavioral event characterization, t, for characterizing the jth sample behavioral event j For characterizing the moment of occurrence of the j-th sample behaviour event, b t Correction quantity, w, for characterizing a first probability density function 1 And w 2 All the model parameters are model parameters for transforming time quantity, and p is used for representing the hyper-parameters of the behavior prediction model;
the conditional strength function defined in the conventional manner is λ * (t)＝exp(v t ·h j +w t (t-t j )+b t ) However, in the calculation for further determining the second prediction probability distribution, the model parameter w for the time quantity is transformed t Can occur in division operations, which are extremely prone to numerical calculation problems, such as when w t When the value is around 0, the divide-by-0 abnormality is extremely likely to occur. And in this application is considered to make use of
And
in place of w t By introducing the exponential function, the problem of numerical value calculation cannot be caused in the subsequent division calculation.
In addition, since w 1 And w 2 Has a value in the range of (- ∞, infinity), so that
More than 0,
Is less than 0. And in exp (v) t ·h j +w t (t-t j )+b t ) In, w t The value range of (c) is (- ∞, + ∞). To make up for
And
the deficiency of two degrees of freedom of the components is two functions
And
as a function of the first probability density.
It should be noted that p is a hyper-parameter of the behavior prediction model, and can be determined through empirical value or experimental value verification.
It can be seen that the first probability density function determined by the present scheme can avoid the occurrence of w in the calculation t By division, i.e. w can be avoided t The calculation is abnormal, so that the accuracy of predicting the occurrence time of the event can be improved.
Step 603 will be explained below.
In this step, the self t is determined from the first probability density function i When there is no second probability density function of any behavior event occurring within a preset time period from the time, the second probability density function may be determined by using the following calculation formula:
wherein S is * (t) for characterizing a second probability density function, λ * (t) characterizing the first probability density function;
step 605 will be explained below.
When the description t is obtained i A first probability density function of occurrence of a behavioral event after the time of day, and from t i After the second probability density function without any behavior event occurring within the preset time period from the moment, the joint probability density function of the first probability density function and the second probability density function is considered to be determined, that is, no behavior event occurs within a certain period of time, and the probability density with the behavior event occurring at the future time t is the second prediction probability distribution. In one possible implementation, the stepsStep 605 may determine the second predicted probability distribution by:
f * (t)＝λ * (t)·S * (t)
wherein f is * (t) for characterizing a second predictive probability distribution, λ * (t) for characterizing a first probability density function, S * (t) is used to characterize the second probability density function.
Therefore, the relationship between the behavior event and the occurrence time of the behavior event can be described through the second prediction probability distribution, and the occurrence time of the behavior event can be predicted accordingly.
As shown in fig. 7, an embodiment of the present specification further provides a risk behavior prediction method, which may include the following steps:
step 701: get the T th p The type identification of at least one behavior event before the moment and the time value corresponding to each behavior event;
step 703: inputting the acquired type identifier of at least one behavior event and the time value corresponding to each behavior event into a behavior prediction model to obtain the Tth behavior event p The type identification predicted value of a behavior event to be predicted after the moment and the time predicted value corresponding to the behavior event to be predicted; the behavior prediction model is obtained by training by using a training method of the behavior prediction model according to any embodiment in the specification;
step 705: searching whether a data set which is consistent with the type identification predicted value and the time predicted value exists in a historical behavior database; the historical behavior database stores at least one data set, and each data set comprises a type identifier of a behavior event and a time value corresponding to the behavior event;
step 707: and if the behavior event is not found, the behavior event to be predicted is a behavior event with risk.
In this embodiment, the tth may be obtained when predicting the risk behavior p Inputting the type identification of the behavior event before the moment and the corresponding time value into a behavior prediction model to obtain T p After the moment of timeAnd identifying the predicted value and the time predicted value of the type of the behavior event to be predicted. Because the type and the occurrence time of the behavior event generally have certain periodicity and regularity, whether a data set consistent with the type identification predicted value and the time predicted value exists or not can be searched from the historical database. If the user behavior is found, the user behavior is in the daily behavior category and is not risky. And if the behavior event is not found, the behavior event to be predicted is a behavior event with risk.
It is readily understood that what should be stored in the historical behavior database is daily behavior that the user does not have a risk. The historical behavior database comprises a plurality of data sets, and each data set comprises a type identifier of a behavior event and a time value corresponding to the behavior event.
It should be noted that when the type identifier of the behavior event and the corresponding time value of each behavior event are input into the behavior prediction model to obtain the predicted value of the type identifier, the type identifier with the highest probability value may be used as the predicted value of the type identifier of the behavior event to be predicted. Whereas for the prediction of the time of occurrence of a behavioral event, the probability distribution f is due to the second prediction * (t) is a probability density function in continuous time, and since the output at the time when the maximum probability density is present is not necessarily reasonable, it is considered that the expectation is taken as t P+1 The prediction of the time further can estimate the expectation by a method of importance sampling. For example, the following calculation procedure may be utilized for t P+1 Estimating the time:
wherein the content of the first and second substances,
is the pair t P+1 The estimated time p (t) is an exponential distribution c.exp (-c (t-t) P ) C) is t in the conditional strength function defined in the conventional manner P The value at the time is c = exp (v) t ·h j +w t (t-t P )+b t )＝exp(v t ·h j +b t ). Then randomly generating N random numbers r 1 ,r 2 ,…,r N And (c) if so, then there is,
as can be seen, according to the above prediction estimation method, when the first probability density function is complex or the numerical solution is not easy to calculate, the time when the behavior event occurs can be quickly predicted by the random sampling algorithm.
As shown in fig. 8, an embodiment of the present specification further provides a training apparatus for a behavior prediction model, including: a first acquisition module 801, a characterization module 802, and a training module 803;
a first obtaining module 801 configured to obtain, from historical behavior data, a type identifier of a sample behavior event and time information corresponding to the sample behavior event; each type of sample behavior event corresponds to a type identifier;
the characterization module 802 is configured to characterize the sample behavior event in a continuous time domain according to the type identifier and the time information of the sample behavior event acquired by the first acquisition module 801 to obtain a behavior event characterization;
and the training module 803 is configured to characterize the training behavior prediction model according to the behavior event obtained by the characterization module 802, so as to optimize the type identifier output by the behavior prediction model and the predicted value of time.
In one possible implementation, when characterizing the sample behavior event in a continuous time domain according to the type identifier and the time information of the sample behavior event, the characterization module 802 is configured to perform the following operations for each sample behavior event:
mapping the type identifier of the current sample behavior event to a preset space to obtain a mapping vector of the current sample behavior event;
acquiring the behavior event representation of the behavior event of the previous sample; when the current sample behavior event is a first sample behavior event, the behavior event of the last sample behavior event is characterized as 0;
and determining the behavior event representation of the current sample behavior event according to the mapping vector of the current sample behavior event, the time information corresponding to the current sample behavior event and the behavior event representation of the previous sample behavior event.
In one possible implementation manner, when determining the behavior event characterization of the current sample behavior event according to the mapping vector of the current sample behavior event, the time information corresponding to the current sample behavior event, and the behavior event characterization of the previous sample behavior event, the characterization module 802 is configured to determine the behavior event characterization of the jth sample behavior event by using the following calculation formula:
h j ＝max{W y *y j +W t *t j +W h *h j-1 +b h ,0}
wherein h is j Behavioral event characterization, y, for characterizing the jth sample behavioral event j Mapping vector for characterizing the jth sample behavior event, t j Is used for characterizing the time information corresponding to the j sample behavior event, h j-1 Behavioral event characterization for characterizing the j-1 st sample behavioral event, W y Coefficient for characterizing a linear transformation of a behavioral event, W t For characterizing coefficients that transform time information linearly, W h Coefficients for characterizing a linear transformation of the characterization of the last sample behavior event, b h The correction amount used to characterize behavioral events, max { \8230;, 0} is used to characterize the nonlinear activation function, reLU.
In one possible implementation, when the behavior prediction model includes a first prediction probability distribution that predicts a type of a behavior event, the training module 803, when characterizing the training behavior prediction model from the behavior event, is configured to perform the following operations:
determining a first prediction probability distribution for predicting the type of the behavior event according to the behavior event representation;
model parameters in the first prediction probability distribution are optimized using a cross entropy loss function.
In one possible implementation, when determining the first prediction probability distribution for predicting the type of the behavioral event according to the behavioral event characterization, the training module 803 is configured to calculate the first prediction probability distribution for predicting the type of the behavioral event by using the following calculation formula:
wherein P (k) is used to characterize a first predictive probability distribution, V k A parameter matrix with a neural network layer corresponding type identifier k for representing a behavior prediction model, h j The behavioral event characterization method is used for characterizing the behavioral event characterization of the jth sample, K is used for characterizing the type number of the behavioral events of the sample, b 1 And b 2 All are corrections for adjusting the predicted values.
In one possible implementation, when the behavior prediction model includes a second prediction probability distribution that predicts the time of occurrence of the behavior event, the training module 803, when characterizing the training behavior prediction model from the behavior event, is configured to perform the following operations:
determining a second prediction probability distribution for predicting the time of occurrence of the behavioral event according to the behavioral event representation;
model parameters in the second predicted probability distribution are optimized using a logarithmic function.
In one possible implementation, the historical behavior data is t i Data prior to the time of day;
the training module 803, in determining a second prediction probability distribution for predicting the time of occurrence of the behavioral event from the behavioral event characterization, is configured to perform the following:
determining t from behavioral event characterization i A first probability density function of the occurrence of a behavioral event after the time of day;
from the first probability density function, determine t i A second probability density function without any behavior event occurring in a preset time period from the moment;
a second predicted probability distribution is determined based on the first probability density function and the second probability density function.
In one possible implementation, the training module 803 determines t based on the behavioral event characterization i When there is a first probability density function of the occurrence of the behavioral event after the time of day, the method is configured to determine the first probability density function using the following equation:
wherein λ is * (t) for characterizing a first probability density function, v t Model parameters for linear transformation of behavioral event characterization, h j Behavioral event characterization, t, for characterizing the jth sample behavioral event j For characterizing the moment of occurrence of the j-th sample behaviour event, b t Correction quantity, w, for characterizing a first probability density function 1 And w 2 All the model parameters are model parameters for transforming time quantity, and p is used for representing the hyper-parameters of the behavior prediction model;
in one possible implementation, the training module 803 determines the self t from the first probability density function i When the second probability density function of any behavior event does not occur within a preset time period from the moment, the method is configured to calculate the second probability density function by using the following calculation formula:
wherein S is * (t) for characterizing a second probability density function, λ * (t) characterizing the first probability density function;
in one possible implementation, the training module 803, in determining the second predictive probability distribution from the first probability density function and the second probability density function, is configured to calculate the second predictive probability distribution using the following calculation:
f * (t)＝λ * (t)·S * (t)
wherein f is * (t) for characterizing a second predictive probability distribution, λ * (t) for characterizing a first probability density function, S * (t) is used to characterize the second probability density function.
In a possible implementation manner, the sample behavior events acquired by the first acquiring module 801 include: transaction actions generated by the user; the transaction behavior comprises: at least one of shopping, transportation, dining, and entertainment.
As shown in fig. 9, the present specification also provides a risk behavior prediction apparatus, including: a second obtaining module 901, an input module 902, a searching module 903 and a determining module 904;
a second obtaining module 901 configured to obtain the Tth p The type identification of at least one behavior event before the moment and the time value corresponding to each behavior event;
an input module 902, configured to input the type identifier of the at least one behavior event acquired by the second acquiring module 901 and the time value corresponding to each behavior event into the behavior prediction model to obtain a tth p The type identification predicted value of a behavior event to be predicted after the moment and the time predicted value corresponding to the behavior event to be predicted; wherein, the behavior prediction model is obtained by training with a training device of the behavior prediction model according to any one of the embodiments;
a searching module 903, configured to search from the historical behavior database whether a data set consistent with the type identifier prediction value obtained by the input module 902 and consistent with the time prediction value exists; the historical behavior database stores at least one data set, and each data set comprises a type identifier of a behavior event and a time value corresponding to the behavior event;
a determining module 904 configured to determine that the behavior event to be predicted is a behavior event with risk if the searching module 903 does not find the behavior event.
The present specification also provides a computer-readable storage medium having stored thereon a computer program which, when executed in a computer, causes the computer to perform the method of any of the embodiments of the specification.
The present specification also provides a computing device comprising a memory having stored therein executable code and a processor that, when executing the executable code, implements the method of any of the embodiments of the specification.
It is to be understood that the configurations illustrated in the embodiments of the present specification do not specifically limit the training device of the behavior prediction model and the risk behavior prediction device. In other embodiments of the specification, the training means and risk behaviour prediction means of the behaviour prediction model may comprise more or fewer components than shown, or some components may be combined, some components may be split, or a different arrangement of components. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.
For the information interaction, execution process, and other contents between the units in the apparatus, the specific contents may refer to the description in the method embodiment of the present specification because the same concept is based on the method embodiment of the present specification, and are not described herein again.
Those skilled in the art will recognize that in one or more of the examples described above, the functions described in this specification can be implemented in hardware, software, hardware, or any combination thereof. When implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium.
The above-mentioned embodiments, the purpose, technical solutions and advantages described in the present specification are further described in detail, it should be understood that the above-mentioned embodiments are only specific embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made on the basis of the technical solutions of the present invention should be included in the scope of the present invention.
Claims (13)
1. The training method of the behavior prediction model comprises the following steps:
acquiring type identification of a sample behavior event and time information corresponding to the sample behavior event from historical behavior data; each type of sample behavior event corresponds to a type identifier;
according to the type identification and the time information of the sample behavior event, characterizing the sample behavior event on a continuous time domain to obtain behavior event characterization;
and training the behavior prediction model according to the behavior event representation so as to optimize the type identification and the predicted value of time output by the behavior prediction model.
2. The method of claim 1, wherein the characterizing the sample behavior event over a continuous time domain according to the type identifier and the time information of the sample behavior event to obtain a behavior event characterization comprises:
for each sample behavioral event, performing:
mapping the type identifier of the current sample behavior event into a preset space to obtain a mapping vector of the current sample behavior event;
acquiring the behavior event representation of the behavior event of the previous sample; when the current sample behavior event is a first sample behavior event, the behavior event of the last sample behavior event is characterized as 0;
and determining the behavior event representation of the current sample behavior event according to the mapping vector of the current sample behavior event, the time information corresponding to the current sample behavior event and the behavior event representation of the last sample behavior event.
3. The method of claim 2, wherein the determining the behavioral event characterization for the current sample behavioral event according to the mapping vector for the current sample behavioral event, the time information corresponding to the current sample behavioral event, and the behavioral event characterization for the last sample behavioral event comprises:
determining the behavior event characterization of the jth sample behavior event by using the following calculation formula:
h j ＝max{W y *y j +W t *t j +W h *h j-1 +b h ,0}
wherein h is j Behavioral event characterization, y, for characterizing the jth sample behavioral event j Mapping vector for characterizing the jth sample behavior event, t j Is used for representing the time information corresponding to the j sample behavior event, h j-1 Behavioral event characterization for characterizing the j-1 st sample behavioral event, W y Coefficient for characterizing a linear transformation of a behavioral event, W t For characterizing coefficients that transform time information linearly, W h Coefficients for characterizing the linear transformation of the characterization of the last sample behavior event, b h The correction amount used to characterize behavioral events, max { \8230;, 0} is used to characterize the nonlinear activation function, reLU.
4. The method of claim 1, wherein the behavioral prediction model includes a first prediction probability distribution that predicts a type of behavioral event;
the training of the behavior prediction model according to the behavior event characterization includes:
determining a first prediction probability distribution for predicting the type of the behavior event according to the behavior event representation;
optimizing model parameters in the first prediction probability distribution using a cross entropy loss function.
5. The method of claim 4, wherein determining a first prediction probability distribution for predicting a type of behavioral event from the behavioral event characterization comprises:
calculating the first prediction probability distribution for predicting the type of the behavior event by using the following calculation formula:
wherein P (k) is used to characterize the first predictive probability distribution, V k A parameter matrix with a type mark k corresponding to a neural network layer for representing the behavior prediction model, h j The behavioral event representation is used for representing the jth sample behavioral event, K is used for representing the type number of the sample behavioral event, b 1 And b 2 All are corrections for adjusting the predicted values.
6. The method of claim 1, wherein the behavioral prediction model includes a second predictive probability distribution predicting times at which behavioral events occur;
the training of the behavioral prediction model from the behavioral event characterization includes:
determining a second prediction probability distribution for predicting the time of occurrence of the behavioral event according to the behavioral event representation;
and optimizing the model parameters in the second prediction probability distribution by using a logarithmic function.
7. The method of claim 6, wherein the historical behavior data is t i Data prior to the time of day;
determining, from the behavioral event characterization, a second prediction probability distribution that predicts a time of occurrence of a behavioral event, comprising:
determining t from the behavioral event characterization i A first probability density function of the occurrence of a behavioral event after the time of day;
according to the firstProbability density function, determined from t i A second probability density function without any behavior event occurring in a preset time period from the moment;
determining the second predictive probability distribution based on the first probability density function and the second probability density function.
8. The method of claim 7, wherein,
determining t from the behavioral event characterization i A first probability density function of behavioral events occurring after the time of day, comprising:
determining the first probability density function using the following calculation:
wherein λ is * (t) characterizing said first probability density function, v t Model parameters for linear transformation of behavioral event characterizations, h j Behavioral event characterization, t, for characterizing the jth sample behavioral event j For characterizing the moment of occurrence of the j-th sample behaviour event, b t Correction quantity, w, for characterizing a first probability density function 1 And w 2 All the model parameters are model parameters for transforming time quantity, and p is used for representing the hyper-parameters of the behavior prediction model;
and/or the presence of a gas in the gas,
the determining from t according to the first probability density function i A second probability density function for no behavior events occurring within a predetermined period of time from the time of day, comprising:
calculating the second probability density function using the following calculation:
wherein S is * (t) characterizing said second probability density function, λ * (t) use forCharacterizing the first probability density function;
and/or the presence of a gas in the gas,
said determining said second predictive probability distribution from said first probability density function and said second probability density function, comprising:
calculating the second predictive probability distribution using the following calculation:
f * (t)＝λ * (t)·S * (t)
wherein f is * (t) for characterizing said second predictive probability distribution, λ * (t) characterizing the first probability density function, S * (t) is used to characterize the second probability density function.
9. The method of any one of claims 1 to 8,
the sample behavioral events include: transaction actions generated by the user; the transaction actions include: at least one of shopping, transportation, dining, and entertainment.
10. A method of risk behavior prediction, comprising:
get the T th p The type identification of at least one behavior event before the moment and the time value corresponding to each behavior event;
inputting the acquired type identifier of the at least one behavior event and the time value corresponding to each behavior event into a behavior prediction model to obtain the Tth p The type identification predicted value of a behavior event to be predicted after the moment and the time predicted value corresponding to the behavior event to be predicted; wherein the behavior prediction model is obtained by training by using the training method of the behavior prediction model according to any one of claims 1 to 9;
searching whether a data set which is consistent with the type identification predicted value and the time predicted value exists in a historical behavior database; at least one data set is stored in the historical behavior database, and each data set comprises a type identifier of a behavior event and a time value corresponding to the behavior event;
and if the behavior event to be predicted is not found, the behavior event to be predicted is a behavior event with risk.
11. The training device of the behavior prediction model comprises: the system comprises a sample data acquisition module, a characterization module and a training module;
the first acquisition module is configured to acquire the type identifier of the sample behavior event and the time information corresponding to the sample behavior event from historical behavior data; each type of sample behavior event corresponds to a type identifier;
the characterization module is configured to characterize the sample behavior event in a continuous time domain according to the type identifier and the time information of the sample behavior event acquired by the first acquisition module to obtain a behavior event characterization;
and the training module is configured to train the behavior prediction model according to the behavior event representation obtained by the representation module so as to optimize the type identifier output by the behavior prediction model and the predicted value of time.
12. A risk-of-behavior prediction apparatus, comprising: the device comprises a second acquisition module, an input module, a search module and a determination module;
the second obtaining module is configured to obtain the Tth p The type identification of at least one behavior event before the moment and the time value corresponding to each behavior event;
the input module is configured to input the type identifier of the at least one behavior event acquired by the second acquisition module and the time value corresponding to each behavior event into the behavior prediction model to obtain a tth behavior p The type identification predicted value of a behavior event to be predicted after the moment and the time predicted value corresponding to the behavior event to be predicted; wherein the behavior prediction model is trained by using the training device of the behavior prediction model according to claim 11;
the searching module is configured to search whether a data set which is consistent with the type identification predicted value obtained by the input module and is consistent with the time predicted value exists in a historical behavior database; at least one data set is stored in the historical behavior database, and each data set comprises a type identifier of a behavior event and a time value corresponding to the behavior event;
the determining module is configured to determine that the behavior event to be predicted is a risky behavior event if the searching module does not search the behavior event.
13. A computing device comprising a memory having executable code stored therein and a processor that, when executing the executable code, implements the method of any of claims 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210757609.8A CN115169551A (en) | 2022-06-30 | 2022-06-30 | Training method of behavior prediction model, risk behavior prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210757609.8A CN115169551A (en) | 2022-06-30 | 2022-06-30 | Training method of behavior prediction model, risk behavior prediction method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115169551A true CN115169551A (en) | 2022-10-11 |
Family
ID=83490125
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210757609.8A Pending CN115169551A (en) | 2022-06-30 | 2022-06-30 | Training method of behavior prediction model, risk behavior prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115169551A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024114338A1 (en) * | 2022-11-29 | 2024-06-06 | 蚂蚁财富(上海)金融信息服务有限公司 | Method and apparatus for training behavior prediction model |
-
2022
- 2022-06-30 CN CN202210757609.8A patent/CN115169551A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024114338A1 (en) * | 2022-11-29 | 2024-06-06 | 蚂蚁财富(上海)金融信息服务有限公司 | Method and apparatus for training behavior prediction model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3586277B1 (en) | Training policy neural networks using path consistency learning | |
| US20190042887A1 (en) | Computer System for Building, Training and Productionizing Machine Learning Models | |
| JP2022548654A (en) | Computer-based system, computer component and computer object configured to implement dynamic outlier bias reduction in machine learning models | |
| US11379726B2 (en) | Finite rank deep kernel learning for robust time series forecasting and regression | |
| Chen et al. | Learning multiple similarities of users and items in recommender systems | |
| CN111461463A (en) | Short-term load prediction method, system and equipment based on TCN-BP | |
| US11977978B2 (en) | Finite rank deep kernel learning with linear computational complexity | |
| US20230306505A1 (en) | Extending finite rank deep kernel learning to forecasting over long time horizons | |
| Wong et al. | Online time-sequence incremental and decremental least squares support vector machines for engine air-ratio prediction | |
| Grob et al. | A recurrent neural network survival model: Predicting web user return time | |
| CN115169551A (en) | Training method of behavior prediction model, risk behavior prediction method and device | |
| Biloš et al. | Modeling temporal data as continuous functions with process diffusion | |
| CN115168721A (en) | User interest recommendation method and system integrating collaborative transformation and temporal perception | |
| US20210342691A1 (en) | System and method for neural time series preprocessing | |
| Guan et al. | Instrumental variable estimation in ordinal probit models with mismeasured predictors | |
| Lei et al. | A novel time-delay neural grey model and its applications | |
| CN116777056A (en) | Method and device for training prediction model and determining material demand | |
| KR20200112495A (en) | Server and method for providing commercial analysis services | |
| CN113435927B (en) | User willingness prediction method, device, equipment and storage medium | |
| CN114708110A (en) | Joint training method and device for continuous guarantee behavior prediction model and electronic equipment | |
| CN112836168A (en) | Resource reservation quantity determining method, device and storage medium | |
| Lai et al. | Attribute‐Associated Neuron Modeling and Missing Value Imputation for Incomplete Data | |
| Liu et al. | Research on the prediction of short time series based on EMD-LSTM | |
| Li et al. | Ethics in Action: Training Reinforcement Learning Agents for Moral Decision-making In Text-based Adventure Games | |
| Miller et al. | Knowledge Enhanced Deep Learning: Application to Pandemic Prediction |
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 |