WO2023279694A1 - 车辆置换预测方法、装置、设备及存储介质 - Google Patents
车辆置换预测方法、装置、设备及存储介质 Download PDFInfo
- Publication number
- WO2023279694A1 WO2023279694A1 PCT/CN2022/071479 CN2022071479W WO2023279694A1 WO 2023279694 A1 WO2023279694 A1 WO 2023279694A1 CN 2022071479 W CN2022071479 W CN 2022071479W WO 2023279694 A1 WO2023279694 A1 WO 2023279694A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- dimensional feature
- feature vector
- vector
- vehicle
- vectors
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 239000013598 vector Substances 0.000 claims abstract description 449
- 238000012549 training Methods 0.000 claims description 88
- 238000012545 processing Methods 0.000 claims description 64
- 230000004927 fusion Effects 0.000 claims description 46
- 238000013507 mapping Methods 0.000 claims description 22
- 239000000284 extract Substances 0.000 claims description 19
- 239000011159 matrix material Substances 0.000 claims description 18
- 230000004913 activation Effects 0.000 claims description 14
- 238000000605 extraction Methods 0.000 claims description 14
- 238000012417 linear regression Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 14
- 238000012937 correction Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 5
- 238000007405 data analysis Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000012216 screening Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 238000007499 fusion processing Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000009849 deactivation Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/08—Insurance
Definitions
- the present application relates to the field of big data, and in particular to a vehicle replacement prediction method, device, equipment and storage medium.
- customer renewal rate is an important business indicator, and when customers replace new cars, they often change insurance companies due to the guidance of 4S stores, resulting in the loss of renewal insurance.
- the main purpose of this application is to solve the technical problem of lack of standardized automatic analysis of relevant data in the vehicle replacement prediction scenario.
- the first aspect of the present application provides a vehicle replacement prediction method, including: obtaining the historical auto insurance information of the vehicle, extracting the i-dimensional feature vector related to vehicle replacement in the historical auto insurance information;
- the first-order learning model fuses each of the i-dimensional feature vectors to obtain the i+1-dimensional feature vector in the historical auto insurance information, and predicts the first probability of vehicle replacement through the i+1-dimensional feature vector, where i is A positive integer greater than or equal to 1; based on the i-dimensional feature vector, use the high-order learning model in the pre-trained model set to extract the j-dimensional feature vector related to vehicle replacement in the historical auto insurance information, and pass the j Dimensional feature vector, predicting the second probability of vehicle replacement, wherein j is a positive integer greater than or equal to i; using the modified model in the pre-training model set to fuse the first probability and the second probability to obtain the fusion probability , and according to the fusion probability, determine the prediction result of the historical vehicle replacement by the pre-training model set; use
- the second aspect of the present application provides a vehicle replacement prediction device, including a memory, a processor, and computer-readable instructions stored in the memory and operable on the processor, and the processor executes the computer-readable instructions.
- the following steps are implemented when reading the instruction: obtain the historical auto insurance information of the vehicle, extract the i-dimensional feature vector related to vehicle replacement in the historical auto insurance information; use the low-order learning model in the preset pre-training model set to The feature vectors are fused to obtain the i+1-dimensional feature vector in the historical auto insurance information, and the first probability of vehicle replacement is predicted by the i+1-dimensional feature vector, wherein, i is a positive integer greater than or equal to 1; based on the The i-dimensional feature vector, using the high-order learning model in the pre-training model set to extract the j-dimensional feature vector related to vehicle replacement in the historical auto insurance information, and predicting the first dimension of the vehicle replacement through the j-dimensional feature vector Two probabilities, wherein, j is a positive integer greater than or equal to i
- the third aspect of the present application provides a computer-readable storage medium, where computer instructions are stored in the computer-readable storage medium, and when the computer instructions are run on the computer, the computer is made to perform the following steps: Obtain the historical auto insurance of the vehicle Information, extracting i-dimensional feature vectors related to vehicle replacement in the historical auto insurance information; using the low-order learning model in the preset pre-training model set to fuse each of the i-dimensional feature vectors to obtain the i-dimensional feature vectors in the historical auto insurance information The i+1-dimensional feature vector, and through the i+1-dimensional feature vector, predict the first probability of vehicle replacement, wherein, i is a positive integer greater than or equal to 1; based on the i-dimensional feature vector, the pre-training model is used The high-order learning model in the set extracts the j-dimensional feature vector related to vehicle replacement in the historical auto insurance information, and predicts the second probability of vehicle replacement through the j-dimensional feature vector, wherein j is a positive integer greater than or equal to i ;
- the fourth aspect of the present application provides a vehicle replacement prediction device, wherein the vehicle replacement prediction device includes: an input module, used to obtain the historical auto insurance information of the vehicle, and extract the i-dimension related to vehicle replacement in the historical auto insurance information Feature vector; a low-order feature fusion module, which is used to fuse each of the i-dimensional feature vectors using a low-order learning model in a preset pre-training model set to obtain the i+1-dimensional feature vector in the historical auto insurance information, And predict the first probability of vehicle replacement through the i+1-dimensional feature vector, wherein, i is a positive integer greater than or equal to 1; the high-order feature extraction module is used to adopt the pre-training model based on the i-dimensional feature vector The high-order learning model in the set extracts the j-dimensional feature vector related to vehicle replacement in the historical auto insurance information, and predicts the second probability of vehicle replacement through the j-dimensional feature vector, wherein j is a positive integer greater than or equal to i ; A determination module, configured to fuse
- the historical auto insurance information of the vehicle is obtained and the i-dimensional feature vector is extracted; the low-order learning model is used to predict the first probability of vehicle replacement, and the high-order learning model is used to predict the second probability of vehicle replacement.
- Both the high-order learning model and the high-order learning model do not need to use artificial feature engineering to process historical auto insurance information, and can learn directly, which can greatly speed up the iteration speed of the model; in addition, the low-order learning model extracts a large number of more specific , low-dimensional i+1-dimensional feature vector, which is more comprehensive when used for feature analysis of vehicle replacement prediction, and the first probability predicted by the low-order learning model is more visualized when representing the probability of car insurance replacement; while the high-order learning model is extracted to obtain Fewer, more abstract, and high-dimensional j-dimensional feature vectors, through higher-order common features related to vehicle replacement, when predicting vehicle replacement, focus more on the basic features of vehicle replacement, and predict through high-level learning models The resulting second probability is more general.
- the prediction result of vehicle replacement is determined.
- the fusion probability combines the first probability of concrete visualization and the second probability of abstract generalization, and characterizes the vehicle replacement from high-dimensional and low-dimensional perspectives. Possibilities, so that the prediction results are more accurate; then the revised model is used to iterate the low-order learning model and the high-order learning model until the low-order learning model and the high-order learning model converge, and a set of vehicle replacement prediction models is obtained.
- Fig. 1 is the schematic diagram of the first embodiment of the vehicle replacement prediction method in the embodiment of the present application
- Fig. 2 is the schematic diagram of the second embodiment of the vehicle replacement prediction method in the embodiment of the present application.
- Fig. 3 is the schematic diagram of the third embodiment of the vehicle replacement prediction method in the embodiment of the present application.
- Fig. 4 is a schematic diagram of an embodiment of the vehicle replacement prediction device in the embodiment of the present application.
- FIG. 5 is a schematic diagram of another embodiment of the vehicle replacement prediction device in the embodiment of the present application.
- Fig. 6 is a schematic diagram of an embodiment of the vehicle replacement prediction device in the embodiment of the present application.
- the embodiment of the present application provides a vehicle replacement prediction method, device, equipment and storage medium, which acquires the historical vehicle insurance information of the vehicle and extracts the i-dimensional feature vector; uses a low-order learning model to predict the first probability of vehicle replacement; uses a high-order learning model The model predicts the second probability of vehicle replacement; based on the fusion probability of the first and second probabilities, the prediction result of vehicle replacement is determined, and the modified model is used to iterate the low-level learning model and the high-level learning model until the low-level learning model and When the high-level learning model converges, a vehicle replacement prediction model set is obtained; the vehicle insurance information of the target vehicle is obtained and input into the vehicle replacement prediction model set to predict the prediction result of the target vehicle replacement.
- the application speeds up the iterative speed and prediction speed of the vehicle replacement prediction model, increases the probability of auto insurance extension when the vehicle is replaced, and reduces the loss of vehicle renewal customers.
- the first embodiment of the vehicle replacement prediction method in the embodiment of the present application includes:
- the subject of execution of the present application may be a vehicle replacement prediction device, and may also be a terminal or a server, which is not specifically limited here.
- the embodiment of the present application is described by taking the server as an execution subject as an example.
- the auto insurance information of the vehicle contains a plurality of data in a coded format, including: basic information of the insured vehicle, historical insurance records, historical accident records, basic information of the applicant, LBS (Location Based Services) behavior data of the applicant, Policyholder extended information, etc. It is also possible to further process the data in the auto insurance information, such as further text processing the POI (Point Of Interests) data of the policyholder LBS, and distinguish the characteristics of the crowd.
- POI Point Of Interests
- the low-order learning model is used to train the i-dimensional feature vector and the low-dimensional feature vector, which can reduce data sparsity and reduce the impact of data noise and redundancy. Increase the scalability of model training; use the high-dimensional learning model to extract the deep features of auto insurance information, so that the results of the model’s prediction of vehicle replacement are more in line with the hidden deep features of auto insurance information;
- the first-order learning model is iterated to solve the problem of feature gradient backpropagation message in the process of model training, the problem of parameter update effectiveness, and the problem of poor model convergence caused by it.
- the i-dimensional feature vector in the auto insurance information is used to represent the probability of the owner's vehicle replacement, the feature sparsity is relatively obvious, so here, the correlation between different i-dimensional feature vectors is used, and the low-order learning model is used for each The i-dimensional feature vector is fused.
- the two-dimensional feature vector is used to represent the correlation between the i-dimensional feature vectors
- the subsequent three-dimensional feature vectors are used to represent the correlation between the two-dimensional feature vectors, and so on.
- the i-dimensional feature vector can be fused by a preset fusion method in the input layer of the low-order learning model, such as a logistic regression method, a k-nearest neighbor ((k-Nearest Neighbor, KNN)) method, a support vector machine , FM (Factorization Machi, factorization machine), etc., are fused in a cyclic fusion manner, wherein, i can be set according to business requirements or scene characteristics, preferably, i ⁇ 4.
- a preset fusion method in the input layer of the low-order learning model such as a logistic regression method, a k-nearest neighbor ((k-Nearest Neighbor, KNN)) method, a support vector machine , FM (Factorization Machi, factorization machine), etc.
- the i-dimensional feature vector when the i-dimensional feature vector is further fused, the i-dimensional feature vector can be extracted and fused by using the convolution kernel of i, so as to realize the generation of the i+1-dimensional feature vector.
- n is the dimension of the i-dimensional feature vector
- ⁇ 0 , ⁇ i , ⁇ ij , and ⁇ ijk are the weight parameters of the initial weight parameter, i-dimensional feature vector, two-dimensional feature vector, and three-dimensional feature vector, respectively .
- a fully connected layer cascading method can be used to pass a DNN (Deep Neural Networks, deep neural network ) regression or classification algorithm for feature fusion and second probability prediction of vehicle replacement.
- DNN Deep Neural Networks, deep neural network
- the low-level learning model training can be a feature vector from one dimension to i-dimensional
- the high-level learning model can train a j-dimensional feature vector higher than i-dimensional.
- the two models predict the probability of vehicle replacement from different dimensions.
- the high-level learning model includes a residual module in the hidden layer of each layer of feature fusion, which preliminarily prevents the disappearance of the gradient in the iteration of the high-level learning model.
- the high-order learning model when used to predict the second probability of vehicle replacement through the j-dimensional feature vector, the following loss function is used to measure the convergence of the high-order learning model:
- W is the weight coefficient of the j-dimensional feature vector in the high-level learning model
- b is the offset vector of the j-dimensional feature vector
- X is the input j-1-dimensional feature vector
- xj is the output j-dimensional feature vector
- ym is Feature vectors of m attribute types in the j-dimensional feature vector.
- the prediction results of the low-level learning model and the high-level learning model are corrected through the correction model.
- the fusion probability of the prediction results of the low-level learning model and the high-level learning model is first calculated, and the two models are further determined.
- the residual between the predicted result fusion and the real predicted value, and the two models are corrected according to the residual value.
- the calculation method of the residual value is as follows:
- r n is the residual value
- a ture is the real predicted value
- the i-dimensional feature vector Xn and the corresponding residual value Input into the modified model processed by the following linear regression equation: according to The value is compared with the preset value, and the low-order learning model and the high-order learning model are determined according to the comparison results to iterate until the two models converge, then the current fusion probability can be output.
- the auto insurance information of the target vehicle that needs to be predicted is input into the vehicle replacement prediction model set, so as to directly predict whether the target vehicle will be replaced, wherein the auto insurance information of the target vehicle is the same as the historical auto insurance information, At least include: basic information of the insured vehicle, historical insurance records, historical accident records, basic information of the policyholder, LBS (Location Based Services) behavior data of the policyholder, extended information of the policyholder, etc.
- LBS Location Based Services
- the historical auto insurance information of the vehicle is obtained and the i-dimensional feature vector is extracted; the low-order learning model is used to predict the first probability of vehicle replacement, and the high-order learning model is used to predict the second probability of vehicle replacement, wherein the low-order learning model Both the model and the high-level learning model do not need to perform artificial feature success on historical auto insurance information, which can greatly speed up the iterative speed of the model; The low-order learning model and the high-order learning model are iterated until the low-order learning model and the high-order learning model converge, and a set of vehicle replacement prediction models is obtained.
- the second embodiment of the vehicle replacement prediction method in the embodiment of the present application includes:
- the preset sparse feature vectors to encode and embed each factor combination, obtain the one-dimensional feature vectors in the historical auto insurance information and input them into the preset pre-training model set, wherein the pre-training model set includes a low-level learning model , a high-level learning model and a revision model;
- the multi-type feature factors in the auto insurance information are divided into multiple attribute categories, and the number of attribute categories is the initial dimension number of the model; firstly, the feature factors of each attribute type are normalized and mapped, such as single heat code (One Hot Vector Mapping) vector mapping; and then cascade the embedded coding layer after the one-hot code input layer, classify the specific feature factors of the same attribute category into the same group, and normalize the normalized mapping of different groups
- the code is mapped to a low-dimensional vector, and the one-dimensional feature vector in the auto insurance information can be obtained to compress the input dimension of the auto insurance information.
- F(x) is the response vector after embedding coding
- S is the unique value of each feature factor.
- M is the preset parameter matrix for encoding embedding.
- the i-dimensional feature vector when fusing the i-dimensional feature vector, is used as the first basic vector of the model to perform the feature fusion process to obtain the two-dimensional feature vector; and when fusing the two-dimensional feature vector, the two-dimensional feature vector Perform the feature fusion process as the first basic vector of the model to obtain a three-dimensional feature vector; and so on until the preset multi-dimensional feature vector is obtained.
- a cross weight matrix is introduced to fuse low-dimensional feature vectors according to the weight matrix.
- the low-level learning model does not need to carry out historical auto insurance information
- the success of artificial features can greatly speed up the iteration speed of the model, reduce the R&D cycle and R&D cost of the model, and quickly predict the result of vehicle replacement after training the model, increase the probability of auto insurance extension during vehicle replacement, and reduce the number of customers who want to renew their vehicle insurance loss.
- the third embodiment of the vehicle replacement prediction method in the embodiment of the present application includes:
- i-dimensional feature vector as the second basic vector, use the high-order learning model in the pre-training model set to perform weighted combination on each of the second basic vectors, obtain multiple weighted combination vectors, and perform non-linearity on each weighted combination vector Mapping processing to obtain a k-dimensional feature vector related to vehicle replacement, where j>k>i;
- the specific weighted combination vector generation includes the following steps:
- the i-dimensional feature vector is further screened through the random deactivation criterion, and the weighted combination obtained from the screening is performed.
- fusion can be performed in the following manner: in, W (j) , b (j) is the weight matrix of the jth layer, the eigenvector of the jth layer, and the bias vector connecting the jth layer and the j+1th layer, ⁇ () is a nonlinear mapping function, and Sigmoid or ReLU( Rectified Linear Unit, linear rectification function).
- a residual structure is introduced to the k-dimensional feature vector that satisfies the jump condition, and the j-dimensional feature vector that should have been generated is replaced by the residual vector.
- the residual vector between the k-2-dimensional feature vector and the k-dimensional feature vector is added according to the set weight through the gating of the two feature vectors, and the weight can be set according to the training process of the high-order learning model.
- the high-order learning model is used to predict the second probability of vehicle replacement, and the high-order features of the historical auto insurance information during vehicle replacement are generalized from a high latitude , which is more conducive to improving the generalization ability and convergence ability of the model;
- the subsequent revision model is used to iterate the low-order learning model and the high-order Selected, improve the accuracy of model prediction, until the model converges, you can quickly predict the result of vehicle replacement in real time, increase the probability of auto insurance extension during vehicle replacement, reduce the loss of vehicle renewal customers, and quickly dispatch staff to follow up.
- An embodiment of the vehicle replacement prediction device in the embodiment of the present application includes:
- the input module 401 is used to obtain the historical auto insurance information of the vehicle, and extracts the i-dimensional feature vector related to vehicle replacement in the historical auto insurance information;
- the low-order feature fusion module 402 is used to fuse each of the i-dimensional feature vectors with the low-order learning model in the preset pre-training model set to obtain the i+1-dimensional feature vector in the historical auto insurance information, and pass i+1-dimensional feature vector, predicting the first probability of vehicle replacement, where i is a positive integer greater than or equal to 1;
- the high-order feature extraction module 403 is configured to use the high-order learning model in the pre-training model set to extract the j-dimensional feature vector related to vehicle replacement in the historical auto insurance information based on the i-dimensional feature vector, and use the The j-dimensional feature vector predicts the second probability of vehicle replacement, wherein j is a positive integer greater than or equal to i;
- a determining module 404 configured to fuse the first probability and the second probability using a modified model in the pre-training model set to obtain a fusion probability, and determine the pair of the pre-training model set according to the fusion probability Prediction results of historical vehicle replacements;
- the residual processing module 405 is configured to use the revised model to calculate the residual value between the predicted result and the actual result of historical vehicle replacement, and use the i-dimensional feature vector and the residual value to calculate the predicted
- the result is processed by linear regression, and the processing result is obtained;
- An iteration module 406 configured to iterate the low-order learning model and the high-order learning model according to the processing results until the low-order learning model and the high-order learning model converge to obtain a vehicle replacement prediction model gather;
- the prediction module 407 is configured to acquire the auto insurance information of the target vehicle, input the auto insurance information of the target vehicle into the vehicle replacement prediction model set, and predict the prediction result of the target vehicle replacement.
- the historical auto insurance information of the vehicle is obtained and the i-dimensional feature vector is extracted; the low-order learning model is used to predict the first probability of vehicle replacement, and the high-order learning model is used to predict the second probability of vehicle replacement, wherein the low-order learning model Both the model and the high-level learning model do not need to perform artificial feature success on historical auto insurance information, which can greatly speed up the iterative speed of the model; The low-order learning model and the high-order learning model are iterated until the low-order learning model and the high-order learning model converge, and a set of vehicle replacement prediction models is obtained.
- FIG. 5 another embodiment of the vehicle replacement prediction device in the embodiment of the present application includes:
- the input module 401 is used to obtain the historical auto insurance information of the vehicle, and extracts the i-dimensional feature vector related to vehicle replacement in the historical auto insurance information;
- the low-order feature fusion module 402 is used to fuse each of the i-dimensional feature vectors with the low-order learning model in the preset pre-training model set to obtain the i+1-dimensional feature vector in the historical auto insurance information, and pass i+1-dimensional feature vector, predicting the first probability of vehicle replacement, where i is a positive integer greater than or equal to 1;
- the high-order feature extraction module 403 is configured to use the high-order learning model in the pre-training model set to extract the j-dimensional feature vector related to vehicle replacement in the historical auto insurance information based on the i-dimensional feature vector, and use the The j-dimensional feature vector predicts the second probability of vehicle replacement, wherein j is a positive integer greater than or equal to i;
- a determining module 404 configured to fuse the first probability and the second probability using a modified model in the pre-training model set to obtain a fusion probability, and determine the pair of the pre-training model set according to the fusion probability Prediction results of historical vehicle replacements;
- the residual processing module 405 is configured to use the revised model to calculate the residual value between the predicted result and the actual result of historical vehicle replacement, and use the i-dimensional feature vector and the residual value to calculate the predicted
- the result is processed by linear regression, and the processing result is obtained;
- An iteration module 406 configured to iterate the low-order learning model and the high-order learning model according to the processing results until the low-order learning model and the high-order learning model converge to obtain a vehicle replacement prediction model gather;
- the prediction module 407 is configured to acquire the auto insurance information of the target vehicle, input the auto insurance information of the target vehicle into the vehicle replacement prediction model set, and predict the prediction result of the target vehicle replacement.
- the input module 401 includes:
- An acquisition unit 4011 configured to acquire historical auto insurance information of the vehicle
- the input unit 4013 is configured to input the i-dimensional feature vector into a preset pre-training model set, wherein the pre-training model set includes a low-level learning model, a high-level learning model and a modified model;
- the low-order feature fusion module 402 includes:
- the low-level feature fusion unit 4021 is configured to use the i-dimensional feature vector as the first basic vector, and use the low-level learning model in the preset pre-training model set to combine every two first basic vectors to obtain multiple vectors Combining; according to the number of vector combinations, establish a cross weight matrix corresponding to the i-dimensional feature vector, and according to the cross weight matrix, sequentially fuse the two first basic vectors in each vector combination to obtain the corresponding i+1 dimension feature vector;
- the first prediction unit 4022 is configured to predict the first probability of vehicle replacement through i+1-dimensional feature vectors, where i is a positive integer greater than or equal to 1.
- the high-order feature extraction module 403 includes:
- the high-order feature extraction unit 4031 is configured to use the i-dimensional feature vector as the second basic vector, and use the high-order learning model in the pre-training model set to perform weighted combination on each of the second basic vectors to obtain multiple Weighting the combined vectors, and performing nonlinear mapping processing on each of the weighted combined vectors to obtain a k-dimensional feature vector related to vehicle replacement, wherein j>k>i; using the k-dimensional feature vector as a new second basic vector , and judge whether the new second basis vector satisfies the preset jump condition; if so, calculate the residual vector corresponding to the k-dimensional feature vector, and use the residual vector as the k+1-dimensional feature vector , using the k+1-dimensional feature vector as a new second basis vector, performing weighted combination and nonlinear mapping processing on each of the new second basis vectors, until the j-dimensional feature vector is obtained; if not satisfied, Then perform weighted combination and non-linear mapping processing on each of the new second basic vectors
- the second prediction unit 4032 is configured to predict the second probability of vehicle replacement through the j-dimensional feature vector, where j is a positive integer greater than or equal to i
- the high-order feature extraction unit 4031 is also used for:
- a weighting coefficient corresponding to the selected second basic vector is determined according to the activation probability, and weighted combination is performed on each of the second basic vectors by using the weighting coefficient to obtain a plurality of weighted combination vectors.
- the low-level learning model after obtaining the historical auto insurance information of the vehicle, by extracting the i-dimensional feature vector, and using the low-level learning model to predict the first probability of vehicle replacement, wherein the low-level learning model does not need to carry out historical auto insurance information
- the success of artificial features can greatly speed up the iteration speed of the model, reduce the R&D cycle and R&D cost of the model, and quickly predict the result of vehicle replacement after training the model, increase the probability of auto insurance extension during vehicle replacement, and reduce the number of customers who want to renew their vehicle insurance
- the high-order learning model is used to predict the second probability of vehicle replacement, and the high-order features of historical auto insurance information during vehicle replacement are generalized from a high latitude, It is more conducive to improving the generalization ability and convergence ability of the model; the modified model is used to iterate the low-level learning model and the high-level learning model, and the low-level and high-level models are respectively
- FIG 4 and Figure 5 above describe the vehicle replacement prediction device in the embodiment of the present application in detail from the perspective of modular functional entities, and the following describes the vehicle replacement prediction device in the embodiment of the present application in detail from the perspective of hardware processing.
- Fig. 6 is a schematic structural diagram of a vehicle replacement prediction device provided by an embodiment of the present application.
- the vehicle replacement prediction device 600 may have relatively large differences due to different configurations or performances, and may include one or more central processing units. , CPU) 610 (eg, one or more processors) and memory 620, and one or more storage media 630 (eg, one or more mass storage devices) for storing application programs 633 or data 632.
- the memory 620 and the storage medium 630 may be temporary storage or persistent storage.
- the program stored in the storage medium 630 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations for the vehicle replacement prediction device 600 .
- the processor 610 may be configured to communicate with the storage medium 630 , and execute a series of instruction operations in the storage medium 630 on the vehicle replacement prediction device 600 .
- the vehicle replacement prediction device 600 may also include one or more power sources 640, one or more wired or wireless network interfaces 650, one or more input and output interfaces 660, and/or, one or more operating systems 631, such as Windows Server , Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that the structure of the vehicle replacement prediction device shown in FIG. Part placement.
- the present application also provides a vehicle replacement prediction device.
- the computer device includes a memory and a processor.
- Computer-readable instructions are stored in the memory.
- the processor executes the steps in the above-mentioned embodiments. The steps of the vehicle replacement prediction method.
- the present application also provides a computer-readable storage medium.
- the computer-readable storage medium may be a non-volatile computer-readable storage medium.
- the computer-readable storage medium may also be a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium, and when the instructions are run on the computer, the computer is made to execute the steps of the vehicle replacement prediction method.
- the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
- the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application.
- the aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (read-only memory, ROM), random access memory (random access memory, RAM), magnetic disk or optical disc and other media that can store program codes. .
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- Finance (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Accounting & Taxation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- General Business, Economics & Management (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
- Traffic Control Systems (AREA)
Abstract
本申请涉及大数据领域,公开了一种车辆置换预测方法、装置、设备及存储介质。该方法包括:获取车辆的历史车险信息并提取i维特征向量;采用低阶学习模型预测车辆置换的第一概率;采用高阶学习模型预测车辆置换的第二概率;基于第一、第二概率的融合概率,确定车辆置换的预测结果,并采用修正模型对低阶学习模型和高阶学习模型进行迭代,直到低阶学习模型和高阶学习模型收敛时,得到车辆置换预测模型集合;获取目标车辆的车险信息并输入车辆置换预测模型集合中,预测目标车辆置换的预测结果。本申请加快了车辆置换预测模型的迭代速度和预测速度,提升车辆置换时的车险延保的概率,降低车辆续保客户的流失。
Description
本申请要求于2021年07月05日提交中国专利局、申请号为202110754151.6、发明名称为“车辆置换预测方法、装置、设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
本申请涉及大数据领域,尤其涉及一种车辆置换预测方法、装置、设备及存储介质。
对于保险行业来说,客户续保率是一个重要的业务指标,而客户在置换新车的时候往往会因4S店的引导而更换保险公司,导致续保的流失。
而目前业内对置换车线索的获取主要依赖于被动触发的方式,且全程由人工进行数据线索跟踪,预测客户车辆置换的可能性,但绝大多数提供线索的数据质量并不高,而且更多取决于调研人员的专业能力,不仅仅是数据本身。发明人意识到,在数据选择方面,客户的保险信息中,存在大量可用于预测受保车辆置换的数据,而在数据分析方面,如果仅采用人工的方式进行数据分析,难以充分挖掘出数据中有关车辆置换的特征,且不同调研人员对数据的分析能力不一,也会影响数据分析结果,故在车辆置换预测场景中,缺少对相关数据进行规范化的自动化分析。
发明内容
本申请的主要目的在于解决在车辆置换预测场景中,缺少对相关数据进行规范化的自动化分析的技术问题。
本申请第一方面提供了一种车辆置换预测方法,包括:获取车辆的历史车险信息,提取所述历史车险信息中与车辆置换相关的i维特征向量;采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量,并通过所述j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数;采用所述预训练模型集合中的修正模型对所述第一概率和所述第二概率融合,得到融合概率,并根据所述融合概率,确定所述预训练模型集合对历史车辆置换的预测结果;采用所述修正模型计算所述预测结果和历史车辆置换的真实结果之间的残差值,通过所述i维特征向量和所述残差值,对所述预测结果进行线性回归处理,得到处理结果;根据所述处理结果对所述低阶学习模型和所述高阶学习模型进行迭代,直到所述低阶学习模型和所述高阶学习模型收敛时,得到车辆置换预测模型集合;获取目标车辆的车险信息,并将所述目标车辆的车险信息输入所述车辆置换预测模型集合中,预测所述目标车辆置换的预测结果。
本申请第二方面提供了一种车辆置换预测设备,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可读指令,所述处理器执行所述计算机可读指令时实现如下步骤:获取车辆的历史车险信息,提取所述历史车险信息中与车辆置换相关的i维特征向量;采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量,并通过所述j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数;采用所述预训练模型集合中的修正模型对所述第一概率和所述第二概率融合,得到融合概率,并根据所述融合概率,确定所述预训练模型集合对历史车辆置换的预测结果;采用所述修正模型计算所述预测结果和历史车辆置换的真实结果之间的残差值,通过所述i维特征向 量和所述残差值,对所述预测结果进行线性回归处理,得到处理结果;根据所述处理结果对所述低阶学习模型和所述高阶学习模型进行迭代,直到所述低阶学习模型和所述高阶学习模型收敛时,得到车辆置换预测模型集合;获取目标车辆的车险信息,并将所述目标车辆的车险信息输入所述车辆置换预测模型集合中,预测所述目标车辆置换的预测结果。
本申请的第三方面提供了一种计算机可读存储介质,所述计算机可读存储介质中存储计算机指令,当所述计算机指令在计算机上运行时,使得计算机执行如下步骤:获取车辆的历史车险信息,提取所述历史车险信息中与车辆置换相关的i维特征向量;采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量,并通过所述j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数;采用所述预训练模型集合中的修正模型对所述第一概率和所述第二概率融合,得到融合概率,并根据所述融合概率,确定所述预训练模型集合对历史车辆置换的预测结果;采用所述修正模型计算所述预测结果和历史车辆置换的真实结果之间的残差值,通过所述i维特征向量和所述残差值,对所述预测结果进行线性回归处理,得到处理结果;根据所述处理结果对所述低阶学习模型和所述高阶学习模型进行迭代,直到所述低阶学习模型和所述高阶学习模型收敛时,得到车辆置换预测模型集合;获取目标车辆的车险信息,并将所述目标车辆的车险信息输入所述车辆置换预测模型集合中,预测所述目标车辆置换的预测结果。
本申请第四方面提供了一种车辆置换预测装置,其中,所述车辆置换预测装置包括:输入模块,用于获取车辆的历史车险信息,提取所述历史车险信息中与车辆置换相关的i维特征向量;低阶特征融合模块,用于采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;高阶特征提取模块,用于基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量,并通过所述j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数;确定模块,用于采用所述预训练模型集合中的修正模型对所述第一概率和所述第二概率融合,得到融合概率,并根据所述融合概率,确定所述预训练模型集合对历史车辆置换的预测结果;残差处理模块,用于采用所述修正模型计算所述预测结果和历史车辆置换的真实结果之间的残差值,通过所述i维特征向量和所述残差值,对所述预测结果进行线性回归处理,得到处理结果;迭代模块,用于根据所述处理结果对所述低阶学习模型和所述高阶学习模型进行迭代,直到所述低阶学习模型和所述高阶学习模型收敛时,得到车辆置换预测模型集合;预测模块,用于获取目标车辆的车险信息,并将所述目标车辆的车险信息输入所述车辆置换预测模型集合中,预测所述目标车辆置换的预测结果。
本申请提供的技术方案中,获取车辆的历史车险信息并提取i维特征向量;采用低阶学习模型预测车辆置换的第一概率,采用高阶学习模型预测车辆置换的第二概率,其中,低阶学习模型和高阶学习模型均不需要采用人工特征工程对历史车险信息进行处理,即可直接学习,可以大大加快模型的迭代速度;另外,低阶学习模型提取得到数量较大、更为具体、低维度的i+1维特征向量,用于车辆置换预测的特征分析时更全面,通过低阶学习模型预测得到的第一概率在表征车险置换概率时更可视化;而高阶学习模型提取得到数量较少、更为抽象、高维度的j维特征向量,通过更高阶的车辆置换相关的共同特征,对车辆置换进行预测时,更聚焦于车辆置换的基础特征,通过高阶学习模型预测得到的第二概率更具概括性。基于第一、第二概率的融合概率,确定车辆置换的预测结果,融合概率结 合了具体可视化的第一概率和抽象概括性的第二概率,从高维度和低维度两个角度表征车辆置换的可能性,使其预测结果更准确;接着采用修正模型对低阶学习模型和高阶学习模型进行迭代,直到低阶学习模型和高阶学习模型收敛时,得到车辆置换预测模型集合,通过修正模型分别对低阶和高阶模型进行修正,逐步提升模型预测的可信度;最后直接获取目标车辆的车险信息并输入车辆置换预测模型集合中,预测目标车辆置换的预测结果,可快速预测车辆置换的结果,对车险信息进行规范化的自动化分析,且分析更全面,得到预测质量一致、准确度较高的车量置换的预测结果。
图1为本申请实施例中车辆置换预测方法的第一个实施例示意图;
图2为本申请实施例中车辆置换预测方法的第二个实施例示意图;
图3为本申请实施例中车辆置换预测方法的第三个实施例示意图;
图4为本申请实施例中车辆置换预测装置的一个实施例示意图;
图5为本申请实施例中车辆置换预测装置的另一个实施例示意图;
图6为本申请实施例中车辆置换预测设备的一个实施例示意图。
本申请实施例提供了一种车辆置换预测方法、装置、设备及存储介质,获取车辆的历史车险信息并提取i维特征向量;采用低阶学习模型预测车辆置换的第一概率;采用高阶学习模型预测车辆置换的第二概率;基于第一、第二概率的融合概率,确定车辆置换的预测结果,并采用修正模型对低阶学习模型和高阶学习模型进行迭代,直到低阶学习模型和高阶学习模型收敛时,得到车辆置换预测模型集合;获取目标车辆的车险信息并输入车辆置换预测模型集合中,预测目标车辆置换的预测结果。本申请加快了车辆置换预测模型的迭代速度和预测速度,提升车辆置换时的车险延保的概率,降低车辆续保客户的流失。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”、“第三”、“第四”等(如果存在)是用于区别类似的对象,而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换,以便这里描述的实施例能够以除了在这里图示或描述的内容以外的顺序实施。此外,术语“包括”或“具有”及其任何变形,意图在于覆盖不排他的包含,例如,包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元,而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
为便于理解,下面对本申请实施例的具体流程进行描述,请参阅图1,本申请实施例中车辆置换预测方法的第一个实施例包括:
101、获取车辆的历史车险信息,提取历史车险信息中与车辆置换相关的i维特征向量;
可以理解的是,本申请的执行主体可以为车辆置换预测装置,还可以是终端或者服务器,具体此处不做限定。本申请实施例以服务器为执行主体为例进行说明。
本实施例中,车辆的车险信息内包含编码格式的多个数据,包括:被保车辆基本信息、历史投保记录、历史出险记录、投保人基础信息、投保人LBS(Location Based Services)行为数据、投保人扩展信息等。还可以对车险信息内的数据进行进一步的加工,比如对投保人LBS的POI(Point Of Interests)数据进行进一步的文本加工,对人群进行特征区别。
本实施例中,通过车辆置换预测模型对车险信息进行训练时,采用低阶学习模型对i维特征向量和低维度特征向量进行训练,可以降低数据稀疏性,减少数据噪声和冗余的影响,增加模型训练的可扩展性;采用高维度学习模型提取车险信息的深层特征,使得模型对车辆置换预测的结果更贴合车险信息的隐含深度特征;最后采用修正模型对低阶学习模型以及高阶学习模型进行迭代,解决模型训练过程中特征梯度反向传播消息问题,参数更 新有效性问题,而导致的模型收敛性差的问题。
102、采用预置预训练模型集合中的低阶学习模型对各i维特征向量进行融合,得到历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;
本实施例中,通过车险信息中的i维特征向量表征车主车辆置换的概率时,其特征稀疏性较为明显,故此处利用不同i维特征向量之间具有关联性,通过低阶学习模型对各i维特征向量进行融合。
比如i=1时,即二维特征向量用于表示i维特征向量之间的关联性,后续三维特征向量用于表示二维特征向量之间的关联性,以此类推。通过特征融合提升特征表征的密集型,同时降低模型的复杂性,以提升模型的泛化能力。
本实施例中,低阶学习模型的输入层中可以通过预置的融合方法对i维特征向量进行融合,比如逻辑回归方法、k近邻((k-Nearest Neighbor,KNN))方法、支持向量机、FM(Factorization Machi,因子分解机)等,以循环融合的方式进行融合,其中,i可以根据业务需求或者场景特性进行设置,优选地,i≤4。
具体的,在对i维特征向量进行进一步的融合时,可以采用i的卷积核对i维特征向量进行特征提取并进行融合,以实现i+1维特征向量的生成。
本实施例中,低阶学习模型依赖每一次特征融合后得到的i+1维特征向量,预测车辆置换的第一概率,具体的,若i=3,则低阶学习模型的表达方式如下所示:
103、基于i维特征向量,采用预训练模型集合中的高阶学习模型提取历史车险信息中与车辆置换相关的j维特征向量,并通过j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数;
本实施例中,在采用高阶学习模型对i维特征向量进行融合,提取车险信息中的j维特征向量时,可以采用全连接层级联的方式,通过一个DNN(Deep Neural Networks,深度神经网络)回归或者分类算法,进行特征融合和车辆置换的第二概率预测。
本实施例中,低阶学习模型训练可以一维至i维的特征向量,高阶学习模型训练高于i维的j维特征向量,两个模型从不同的维度对车辆置换的概率进行预测,此处高阶学习模型在每层特征融合的隐藏层中夹杂进残差模块,初步预防高阶学习模型迭代中梯度的消失。
另外,在采用高阶学习模型通过j维特征向量预测车辆置换的第二概率时,采用以下损失函数度量高阶学习模型的收敛性:
其中,W为高阶学习模型中j维特征向量的权重系数,b为j维特征向量的偏移向量,X为输入的j-1维特征向量,xj为输出的j维特征向量,ym为j维特征向量中的m个属性类型的特征向量。
104、采用预训练模型集合中的修正模型对第一概率和第二概率融合,得到融合概率,并根据融合概率,确定预训练模型集合对历史车辆置换的预测结果;
本实施例中,通过修正模型对低阶学习模型和高阶学习模型的预测结果进行修正,此 处先计算低阶学习模型和高阶学习模型预测结果的融合概率,并进一步确定两个模型的预测结果融合后与真实预测值之间的残差,并根据残差值对两个模型进行修正。具体的,残差值计算方式如下所示:
105、采用修正模型计算预测结果和历史车辆置换的真实结果之间的残差值,通过i维特征向量和残差值,对预测结果进行线性回归处理,得到处理结果;
106、根据处理结果对低阶学习模型和高阶学习模型进行迭代,直到低阶学习模型和高阶学习模型收敛时,得到车辆置换预测模型集合;
本实施例中,将i维特征向量Xn和对应的残差值
输入修正模型中,通过以下线性回归方程进行处理:
根据
的值与预设值进行对比,根据对比结果确定低阶学习模型和高阶学习模型进行迭代,直到两模型收敛,则可以输出当前的融合概率。
107、获取目标车辆的车险信息,并将目标车辆的车险信息输入车辆置换预测模型集合中,预测目标车辆置换的预测结果。
本实施例中,最后将需要进行预测的目标车辆的车险信息输入车辆置换预测模型集合中,即可直接预测出目标车辆是否会被置换,其中,目标车辆的车险信息中跟历史车险信息相同,至少包括:被保车辆基本信息、历史投保记录、历史出险记录、投保人基础信息、投保人LBS(Location Based Services)行为数据、投保人扩展信息等。
本申请实施例中,获取车辆的历史车险信息并提取i维特征向量;采用低阶学习模型预测车辆置换的第一概率,采用高阶学习模型预测车辆置换的第二概率,其中,低阶学习模型和高阶学习模型均不需要对历史车险信息进行人工特征成功,可以大大加快模型的迭代速度;基于第一、第二概率的融合概率,确定车辆置换的预测结果,并采用修正模型对低阶学习模型和高阶学习模型进行迭代,直到低阶学习模型和高阶学习模型收敛时,得到车辆置换预测模型集合,此处通过修正模型分别对低阶和高阶模型进行修正,逐步将高换车概率的用户甄选出来,提升模型预测的准确度;最后直接获取目标车辆的车险信息并输入车辆置换预测模型集合中,预测目标车辆置换的预测结果,可快速预测车辆置换的结果,提升车辆置换时的车险延保的概率,降低车辆续保客户的流失。
请参阅图2,本申请实施例中车辆置换预测方法的第二个实施例包括:
201、获取车辆的历史车险信息,当i=1时,提取历史车险信息中与车辆置换相关的多个特征因子和各特征因子对应的属性类别,并根据属性类别,对各特征因子进行分组处理,得到多个因子组合;
202、采用预置稀疏特征向量对各因子组合进行编码嵌入,得到历史车险信息中的一维特征向量并输入预置的预训练模型集合中,其中,预训练模型模型集合包括一个低阶学习模型、一个高阶学习模型和一个修正模型;
本实施例中,将车险信息中多类型的特征因子分为多个属性类别,属性类别的数量即为模型的初始维度数量;首先将各属性类型的特征因子进行归一化映射,比如独热码(One Hot Vector Mapping)向量映射;然后在独热码输入层后级联嵌入编码层,将具体相同属性类别的特征因子归入同一个分组中,并将不同分组的归一化映射后的编码映射到低维向量,即可得到车险信息中的一维特征向量,以压缩车险信息的输入维度。
具体的,一维特征向量的编码嵌入可以通过以下公式进行处理:F(x)=f(S,M),其 中,F(x)是嵌入编码后的响应向量,S为各特征因子的独热码向量,M为编码嵌入的预置参数矩阵。
203、将i维特征向量作为第一基础向量,采用预置预训练模型集合中的低阶学习模型对每两个第一基础向量进行组合,得到多个向量组合;
204、根据向量组合的数量,建立i维特征向量对应的交叉权重矩阵,并根据交叉权重矩阵,依次对各向量组合中的两个第一基础向量进行融合,得到对应的i+1维特征向量;
本实施例中,在融合i维特征向量时,将i维特征向量作为模型的第一基础向量执行特征融合流程,得到二维特征向量;而在融合二维特征向量时,将二维特征向量作为模型的第一基础向量执行特征融合流程,得到三维特征向量;以此类推,直到得到预设的多维特征向量时停止。
具体的,在低阶学习模型处理低维度特征向量时,容易出现特征稀疏的情况,导致特征无法聚焦,故引入交叉权重矩阵,根据权重矩阵对低纬度特征向量进行融合。在对第一基础向量进行融合时,为每个第一基础向量Xi=(x1,x2,……,xi)引入一个辅助向量Vi=(vi1,vi2,……vin),根据向量组合的数量i*n,通过Vn=(V1+V2+……+Vi)计算出各基础向量之间对应辅助向量的交叉权重矩阵,比如对于基础向量x1和x2的向量组合,则通过交叉权重矩阵=v1*V2T。
205、通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;
206、基于i维特征向量,采用预训练模型集合中的高阶学习模型提取历史车险信息中与车辆置换相关的j维特征向量,并通过j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数;
207、采用预训练模型集合中的修正模型对第一概率和第二概率融合,得到融合概率,并根据融合概率,确定预训练模型集合对历史车辆置换的预测结果;
208、采用修正模型计算预测结果和历史车辆置换的真实结果之间的残差值,通过i维特征向量和残差值,对预测结果进行线性回归处理,得到处理结果;
209、根据处理结果对低阶学习模型和高阶学习模型进行迭代,直到低阶学习模型和高阶学习模型收敛时,得到车辆置换预测模型集合;
210、获取目标车辆的车险信息,并将目标车辆的车险信息输入车辆置换预测模型集合中,预测目标车辆置换的预测结果。
本申请实施例中,在获取得到车辆的历史车险信息后,通过提取i维特征向量,并采用低阶学习模型预测车辆置换的第一概率,其中,低阶学习模型不需要对历史车险信息进行人工特征成功,可以大大加快模型的迭代速度,降低模型的研发周期和研发成本,在训练得到模型后可快速预测车辆置换的结果,提升车辆置换时的车险延保的概率,降低车辆续保客户的流失。
请参阅图3,本申请实施例中车辆置换预测方法的第三个实施例包括:
301、获取车辆的历史车险信息,提取历史车险信息中与车辆置换相关的i维特征向量;
302、采用预置预训练模型集合中的低阶学习模型对各i维特征向量进行融合,得到历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;
303、将i维特征向量作为第二基础向量,采用预训练模型集合中的高阶学习模型对各第二基础向量进行加权组合,得到多个加权组合向量,并对各加权组合向量进行非线性映射处理,得到与车辆置换相关k维特征向量,其中,j>k>i;
本实施例中,具体的加权组合向量生成包括以下步骤:
(1)将i维特征向量作为第二基础向量,并根据各第二基础向量的预置激活概率,采用预训练模型集合中的高阶学习模型随机选取多个第二基础向量;
(2)根据激活概率确定选取的第二基础向量对应的加权系数,并采用加权系数对各第二基础向量进行加权组合,得到多个加权组合向量。
本实施例中,为了防止高阶学习模型在训练过程中出现过拟合现象,通过随机失活基准进一步对i维特征向量进行筛选,并将筛选得到的加权组合进行。在加权组合过程中,通过预先设置的激活概率,计算每一个i维特征向量的筛选期望值:E=p*X,其中,E为i维特征向量的筛选期望值,p为激活概率;然后根据期望值筛选出一部分的i维特征向量,再以p进一步从预设的加权系数集合中随机筛选每一个i维特征向量的加权系数,对各i维特征向量进行加权组合,即可得到多个加权组合向量。
304、将k维特征向量作为新的第二基础向量,并判断新的第二基础向量是否满足预置跳变条件;
305、若满足,则计算k维特征向量对应的残差向量,并将残差向量作为k+1维特征向量,将k+1维特征向量作为新的第二基础向量,对各新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止;
306、若不满足,则对各新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止;
本实施例中,对于j维特征向量,可以通过以下方式进行融合:
其中,
W
(j)、
b
(j)分别为第j层的权重矩阵、第j层的特征向量、连接第j层和第j+1层的偏置向量,θ()为非线性映射函数,可以采用Sigmoid或者ReLU(Rectified Linear Unit,线性整流函数)。
另外,还对满足跳变条件的k维特征向量引入残差结构,通过残差向量代替本应生成的j维特征向量。具体的,跳变条件可以设置为k=3N,即每输出两层正常的加权组合和非线性处理的j维特征向量后,加入一个残差结构。
具体的,如果跳变条件设置为k=3N,当当前的k维特征向量满足跳变条件时,计算k-2维特征向量和k维特征向量之间的残差向量,并将计算得到的残差向量作为k维特征向量提取后输出的k=1维特征向量。
其中,k-2维特征向量和k维特征向量之间的残差向量通过两个特征向量的选通按照设定的权重进行相加,而权重可以根据高阶学习模型的训练过程进行设置。
307、通过j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数;
308、采用预训练模型集合中的修正模型对第一概率和第二概率融合,得到融合概率,并根据融合概率,确定预训练模型集合对历史车辆置换的预测结果;
309、采用修正模型计算预测结果和历史车辆置换的真实结果之间的残差值,通过i维特征向量和残差值,对预测结果进行线性回归处理,得到处理结果;
310、根据处理结果对低阶学习模型和高阶学习模型进行迭代,直到低阶学习模型和高阶学习模型收敛时,得到车辆置换预测模型集合;
311、获取目标车辆的车险信息,并将目标车辆的车险信息输入车辆置换预测模型集合中,预测目标车辆置换的预测结果。
本申请实施例中,在获取车辆的历史车险信息并提取i维特征向量后,采用高阶学习模型预测车辆置换的第二概率,从高纬度上泛化车辆置换时历史车险信息的高阶特征,更有利于提升模型的泛化能力和收敛能力;后续采用修正模型对低阶学习模型和高阶学习模型进行迭代,分别对低阶和高阶模型进行修正,逐步将高换车概率的用户甄选出来,提升模型预测的准确度,直到模型收敛,即可实时快速预测车辆置换的结果,提升车辆置换时 的车险延保的概率,降低车辆续保客户的流失,迅速派遣工作人员跟进。
上面对本申请实施例中车辆置换预测方法进行了描述,下面对本申请实施例中车辆置换预测装置进行描述,请参阅图4,本申请实施例中车辆置换预测装置一个实施例包括:
输入模块401,用于获取车辆的历史车险信息,提取所述历史车险信息中与车辆置换相关的i维特征向量;
低阶特征融合模块402,用于采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;
高阶特征提取模块403,用于基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量,并通过所述j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数;
确定模块404,用于采用所述预训练模型集合中的修正模型对所述第一概率和所述第二概率融合,得到融合概率,并根据所述融合概率,确定所述预训练模型集合对历史车辆置换的预测结果;
残差处理模块405,用于采用所述修正模型计算所述预测结果和历史车辆置换的真实结果之间的残差值,通过所述i维特征向量和所述残差值,对所述预测结果进行线性回归处理,得到处理结果;
迭代模块406,用于根据所述处理结果对所述低阶学习模型和所述高阶学习模型进行迭代,直到所述低阶学习模型和所述高阶学习模型收敛时,得到车辆置换预测模型集合;
预测模块407,用于获取目标车辆的车险信息,并将所述目标车辆的车险信息输入所述车辆置换预测模型集合中,预测所述目标车辆置换的预测结果。
本申请实施例中,获取车辆的历史车险信息并提取i维特征向量;采用低阶学习模型预测车辆置换的第一概率,采用高阶学习模型预测车辆置换的第二概率,其中,低阶学习模型和高阶学习模型均不需要对历史车险信息进行人工特征成功,可以大大加快模型的迭代速度;基于第一、第二概率的融合概率,确定车辆置换的预测结果,并采用修正模型对低阶学习模型和高阶学习模型进行迭代,直到低阶学习模型和高阶学习模型收敛时,得到车辆置换预测模型集合,此处通过修正模型分别对低阶和高阶模型进行修正,逐步将高换车概率的用户甄选出来,提升模型预测的准确度;最后直接获取目标车辆的车险信息并输入车辆置换预测模型集合中,预测目标车辆置换的预测结果,可快速预测车辆置换的结果,提升车辆置换时的车险延保的概率,降低车辆续保客户的流失。
请参阅图5,本申请实施例中车辆置换预测装置的另一个实施例包括:
输入模块401,用于获取车辆的历史车险信息,提取所述历史车险信息中与车辆置换相关的i维特征向量;
低阶特征融合模块402,用于采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;
高阶特征提取模块403,用于基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量,并通过所述j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数;
确定模块404,用于采用所述预训练模型集合中的修正模型对所述第一概率和所述第二概率融合,得到融合概率,并根据所述融合概率,确定所述预训练模型集合对历史车辆置换的预测结果;
残差处理模块405,用于采用所述修正模型计算所述预测结果和历史车辆置换的真实结果之间的残差值,通过所述i维特征向量和所述残差值,对所述预测结果进行线性回归处理,得到处理结果;
迭代模块406,用于根据所述处理结果对所述低阶学习模型和所述高阶学习模型进行迭代,直到所述低阶学习模型和所述高阶学习模型收敛时,得到车辆置换预测模型集合;
预测模块407,用于获取目标车辆的车险信息,并将所述目标车辆的车险信息输入所述车辆置换预测模型集合中,预测所述目标车辆置换的预测结果。
具体的,所述输入模块401包括:
获取单元4011,用于获取车辆的历史车险信息;
提取单元4012,用于当i=1时,提取所述历史车险信息中与车辆置换相关的多个特征因子和各所述特征因子对应的属性类别,并根据所述属性类别,对各所述特征因子进行分组处理,得到多个因子组合;采用预置稀疏特征向量对各所述因子组合进行编码嵌入,得到所述历史车险信息中的一维特征向量;
输入单元4013,用于将所述i维特征向量输入预置的预训练模型集合中,其中,所述预训练模型模型集合包括一个低阶学习模型、一个高阶学习模型和一个修正模型;
具体的,所述低阶特征融合模块402包括:
低阶特征融合单元4021,用于将所述i维特征向量作为第一基础向量,采用预置预训练模型集合中的低阶学习模型对每两个第一基础向量进行组合,得到多个向量组合;根据所述向量组合的数量,建立i维特征向量对应的交叉权重矩阵,并根据所述交叉权重矩阵,依次对各所述向量组合中的两个第一基础向量进行融合,得到对应的i+1维特征向量;
第一预测单元4022,用于通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数。
具体的,所述高阶特征提取模块403包括:
高阶特征提取单元4031,用于将所述i维特征向量作为第二基础向量,采用所述预训练模型集合中的高阶学习模型对各所述第二基础向量进行加权组合,得到多个加权组合向量,并对各所述加权组合向量进行非线性映射处理,得到与车辆置换相关k维特征向量,其中,j>k>i;将所述k维特征向量作为新的第二基础向量,并判断所述新的第二基础向量是否满足预置跳变条件;若满足,则计算所述k维特征向量对应的残差向量,并将所述残差向量作为k+1维特征向量,将所述k+1维特征向量作为新的第二基础向量,对各所述新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止;若不满足,则对各所述新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止。
第二预测单元4032,用于通过所述j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数
具体的,所述高阶特征提取单元4031还用于:
将所述i维特征向量作为第二基础向量,并根据各所述第二基础向量的预置激活概率,采用所述预训练模型集合中的高阶学习模型随机选取多个第二基础向量;
根据所述激活概率确定选取的第二基础向量对应的加权系数,并采用所述加权系数对各所述第二基础向量进行加权组合,得到多个加权组合向量。
本申请实施例中,在获取得到车辆的历史车险信息后,通过提取i维特征向量,并采用低阶学习模型预测车辆置换的第一概率,其中,低阶学习模型不需要对历史车险信息进行人工特征成功,可以大大加快模型的迭代速度,降低模型的研发周期和研发成本,在训练得到模型后可快速预测车辆置换的结果,提升车辆置换时的车险延保的概率,降低车辆续保客户的流失;另外,在获取车辆的历史车险信息并提取i维特征向量后,采用高阶学 习模型预测车辆置换的第二概率,从高纬度上泛化车辆置换时历史车险信息的高阶特征,更有利于提升模型的泛化能力和收敛能力;后续采用修正模型对低阶学习模型和高阶学习模型进行迭代,分别对低阶和高阶模型进行修正,逐步将高换车概率的用户甄选出来,提升模型预测的准确度,直到模型收敛,即可实时快速预测车辆置换的结果,提升车辆置换时的车险延保的概率,降低车辆续保客户的流失,迅速派遣工作人员跟进。
上面图4和图5从模块化功能实体的角度对本申请实施例中的车辆置换预测装置进行详细描述,下面从硬件处理的角度对本申请实施例中车辆置换预测设备进行详细描述。
图6是本申请实施例提供的一种车辆置换预测设备的结构示意图,该车辆置换预测设备600可因配置或性能不同而产生比较大的差异,可以包括一个或一个以上处理器(central processing units,CPU)610(例如,一个或一个以上处理器)和存储器620,一个或一个以上存储应用程序633或数据632的存储介质630(例如一个或一个以上海量存储设备)。其中,存储器620和存储介质630可以是短暂存储或持久存储。存储在存储介质630的程序可以包括一个或一个以上模块(图示没标出),每个模块可以包括对车辆置换预测设备600中的一系列指令操作。更进一步地,处理器610可以设置为与存储介质630通信,在车辆置换预测设备600上执行存储介质630中的一系列指令操作。
车辆置换预测设备600还可以包括一个或一个以上电源640,一个或一个以上有线或无线网络接口650,一个或一个以上输入输出接口660,和/或,一个或一个以上操作系统631,例如Windows Serve,Mac OS X,Unix,Linux,FreeBSD等等。本领域技术人员可以理解,图6示出的车辆置换预测设备结构并不构成对车辆置换预测设备的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
本申请还提供一种车辆置换预测设备,所述计算机设备包括存储器和处理器,存储器中存储有计算机可读指令,计算机可读指令被处理器执行时,使得处理器执行上述各实施例中的所述车辆置换预测方法的步骤。
本申请还提供一种计算机可读存储介质,该计算机可读存储介质可以为非易失性计算机可读存储介质,该计算机可读存储介质也可以为易失性计算机可读存储介质,所述计算机可读存储介质中存储有指令,当所述指令在计算机上运行时,使得计算机执行所述车辆置换预测方法的步骤。
所属领域的技术人员可以清楚地了解到,为描述的方便和简洁,上述描述的系统,装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器(read-only memory,ROM)、随机存取存储器(random access memory,RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述,以上实施例仅用以说明本申请的技术方案,而非对其限制;尽管参照前述实施例对本申请进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围。
Claims (20)
- 一种车辆置换预测方法,其中,所述车辆置换预测方法包括:获取车辆的历史车险信息,提取所述历史车险信息中与车辆置换相关的i维特征向量;采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量,并通过所述j维特征向量,预测车辆置换的第二概率,其中,j为大于等于i的正整数;采用所述预训练模型集合中的修正模型对所述第一概率和所述第二概率融合,得到融合概率,并根据所述融合概率,确定所述预训练模型集合对车辆置换的预测结果;采用所述修正模型计算所述预测结果和车辆置换的真实结果之间的残差值,通过所述i维特征向量和所述残差值,对所述预测结果进行线性回归处理,得到处理结果;根据所述处理结果对所述低阶学习模型和所述高阶学习模型进行迭代,直到所述低阶学习模型和所述高阶学习模型收敛时,得到车辆置换预测模型集合;获取目标车辆的车险信息,并将所述目标车辆的车险信息输入所述车辆置换预测模型集合中,预测所述目标车辆置换的预测结果。
- 根据权利要求1所述的车辆置换预测方法,其中,所述提取所述历史车险信息中与车辆置换相关的i维特征向量包括:当i=1时,提取所述历史车险信息中与车辆置换相关的多个特征因子和各所述特征因子对应的属性类别,并根据所述属性类别,对各所述特征因子进行分组处理,得到多个因子组合;采用预置稀疏特征向量对各所述因子组合进行编码嵌入,得到所述历史车险信息中的一维特征向量。
- 根据权利要求1所述的车辆置换预测方法,其中,所述采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量包括:将所述i维特征向量作为第一基础向量,采用预置预训练模型集合中的低阶学习模型对每两个第一基础向量进行组合,得到多个向量组合;根据所述向量组合的数量,建立i维特征向量对应的交叉权重矩阵,并根据所述交叉权重矩阵,依次对各所述向量组合中的两个第一基础向量进行融合,得到对应的i+1维特征向量。
- 根据权利要求1所述的车辆置换预测方法,其中,所述基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量包括:将所述i维特征向量作为第二基础向量,采用所述预训练模型集合中的高阶学习模型对各所述第二基础向量进行加权组合,得到多个加权组合向量,并对各所述加权组合向量进行非线性映射处理,得到与车辆置换相关的k维特征向量,其中,j>k>i;将所述k维特征向量作为新的第二基础向量,并判断所述新的第二基础向量是否满足预置跳变条件;若满足,则计算所述k维特征向量对应的残差向量,并将所述残差向量作为k+1维特征向量,将所述k+1维特征向量作为新的第二基础向量,对各所述新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止;若不满足,则对各所述新的第二基础向量进行加权组合以及非线性映射处理,直到得 到j维特征向量时停止。
- 根据权利要求4所述的车辆置换预测方法,其中,所述将所述i维特征向量作为第二基础向量,采用所述预训练模型集合中的高阶学习模型对各所述第二基础向量进行加权组合,得到多个加权组合向量包括:将所述i维特征向量作为第二基础向量,并根据各所述第二基础向量的预置激活概率,采用所述预训练模型集合中的高阶学习模型随机选取多个第二基础向量;根据所述激活概率确定选取的第二基础向量对应的加权系数,并采用所述加权系数对各所述第二基础向量进行加权组合,得到多个加权组合向量。
- 一种车辆置换预测设备,其中,所述车辆置换预测设备包括:存储器和至少一个处理器,所述存储器中存储有指令;所述至少一个处理器调用所述存储器中的所述指令,以使得所述车辆置换预测设备执行如下所述的车辆置换预测方法:获取车辆的历史车险信息,提取所述历史车险信息中与车辆置换相关的i维特征向量;采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量,并通过所述j维特征向量,预测车辆置换的第二概率,其中,j为大于等于i的正整数;采用所述预训练模型集合中的修正模型对所述第一概率和所述第二概率融合,得到融合概率,并根据所述融合概率,确定所述预训练模型集合对车辆置换的预测结果;采用所述修正模型计算所述预测结果和车辆置换的真实结果之间的残差值,通过所述i维特征向量和所述残差值,对所述预测结果进行线性回归处理,得到处理结果;根据所述处理结果对所述低阶学习模型和所述高阶学习模型进行迭代,直到所述低阶学习模型和所述高阶学习模型收敛时,得到车辆置换预测模型集合;获取目标车辆的车险信息,并将所述目标车辆的车险信息输入所述车辆置换预测模型集合中,预测所述目标车辆置换的预测结果。
- 根据权利要求6所述的车辆置换预测设备,其中,所述提取所述历史车险信息中与车辆置换相关的i维特征向量包括:当i=1时,提取所述历史车险信息中与车辆置换相关的多个特征因子和各所述特征因子对应的属性类别,并根据所述属性类别,对各所述特征因子进行分组处理,得到多个因子组合;采用预置稀疏特征向量对各所述因子组合进行编码嵌入,得到所述历史车险信息中的一维特征向量。
- 根据权利要求6所述的车辆置换预测设备,其中,所述采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量包括:将所述i维特征向量作为第一基础向量,采用预置预训练模型集合中的低阶学习模型对每两个第一基础向量进行组合,得到多个向量组合;根据所述向量组合的数量,建立i维特征向量对应的交叉权重矩阵,并根据所述交叉权重矩阵,依次对各所述向量组合中的两个第一基础向量进行融合,得到对应的i+1维特征向量。
- 根据权利要求6所述的车辆置换预测设备,其中,所述基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维 特征向量包括:将所述i维特征向量作为第二基础向量,采用所述预训练模型集合中的高阶学习模型对各所述第二基础向量进行加权组合,得到多个加权组合向量,并对各所述加权组合向量进行非线性映射处理,得到与车辆置换相关的k维特征向量,其中,j>k>i;将所述k维特征向量作为新的第二基础向量,并判断所述新的第二基础向量是否满足预置跳变条件;若满足,则计算所述k维特征向量对应的残差向量,并将所述残差向量作为k+1维特征向量,将所述k+1维特征向量作为新的第二基础向量,对各所述新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止;若不满足,则对各所述新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止。
- 根据权利要求9所述的车辆置换预测设备,其中,所述将所述i维特征向量作为第二基础向量,采用所述预训练模型集合中的高阶学习模型对各所述第二基础向量进行加权组合,得到多个加权组合向量包括:将所述i维特征向量作为第二基础向量,并根据各所述第二基础向量的预置激活概率,采用所述预训练模型集合中的高阶学习模型随机选取多个第二基础向量;根据所述激活概率确定选取的第二基础向量对应的加权系数,并采用所述加权系数对各所述第二基础向量进行加权组合,得到多个加权组合向量。
- 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指令,其中,所述指令被处理器执行时实现如下所述的车辆置换预测方法:获取车辆的历史车险信息,提取所述历史车险信息中与车辆置换相关的i维特征向量;采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量,并通过所述j维特征向量,预测车辆置换的第二概率,其中,j为大于等于i的正整数;采用所述预训练模型集合中的修正模型对所述第一概率和所述第二概率融合,得到融合概率,并根据所述融合概率,确定所述预训练模型集合对车辆置换的预测结果;采用所述修正模型计算所述预测结果和车辆置换的真实结果之间的残差值,通过所述i维特征向量和所述残差值,对所述预测结果进行线性回归处理,得到处理结果;根据所述处理结果对所述低阶学习模型和所述高阶学习模型进行迭代,直到所述低阶学习模型和所述高阶学习模型收敛时,得到车辆置换预测模型集合;获取目标车辆的车险信息,并将所述目标车辆的车险信息输入所述车辆置换预测模型集合中,预测所述目标车辆置换的预测结果。
- 根据权利要求11所述的计算机可读存储介质,其中,所述提取所述历史车险信息中与车辆置换相关的i维特征向量包括:当i=1时,提取所述历史车险信息中与车辆置换相关的多个特征因子和各所述特征因子对应的属性类别,并根据所述属性类别,对各所述特征因子进行分组处理,得到多个因子组合;采用预置稀疏特征向量对各所述因子组合进行编码嵌入,得到所述历史车险信息中的一维特征向量。
- 根据权利要求11所述的计算机可读存储介质,其中,所述采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维 特征向量包括:将所述i维特征向量作为第一基础向量,采用预置预训练模型集合中的低阶学习模型对每两个第一基础向量进行组合,得到多个向量组合;根据所述向量组合的数量,建立i维特征向量对应的交叉权重矩阵,并根据所述交叉权重矩阵,依次对各所述向量组合中的两个第一基础向量进行融合,得到对应的i+1维特征向量。
- 根据权利要求11所述的计算机可读存储介质,其中,所述基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量包括:将所述i维特征向量作为第二基础向量,采用所述预训练模型集合中的高阶学习模型对各所述第二基础向量进行加权组合,得到多个加权组合向量,并对各所述加权组合向量进行非线性映射处理,得到与车辆置换相关的k维特征向量,其中,j>k>i;将所述k维特征向量作为新的第二基础向量,并判断所述新的第二基础向量是否满足预置跳变条件;若满足,则计算所述k维特征向量对应的残差向量,并将所述残差向量作为k+1维特征向量,将所述k+1维特征向量作为新的第二基础向量,对各所述新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止;若不满足,则对各所述新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止。
- 根据权利要求14所述的计算机可读存储介质,其中,所述将所述i维特征向量作为第二基础向量,采用所述预训练模型集合中的高阶学习模型对各所述第二基础向量进行加权组合,得到多个加权组合向量包括:将所述i维特征向量作为第二基础向量,并根据各所述第二基础向量的预置激活概率,采用所述预训练模型集合中的高阶学习模型随机选取多个第二基础向量;根据所述激活概率确定选取的第二基础向量对应的加权系数,并采用所述加权系数对各所述第二基础向量进行加权组合,得到多个加权组合向量。
- 一种车辆置换预测装置,其中,所述车辆置换预测装置包括:输入模块,用于获取车辆的历史车险信息,提取所述历史车险信息中与车辆置换相关的i维特征向量;低阶特征融合模块,用于采用预置预训练模型集合中的低阶学习模型对各所述i维特征向量进行融合,得到所述历史车险信息中的i+1维特征向量,并通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数;高阶特征提取模块,用于基于所述i维特征向量,采用所述预训练模型集合中的高阶学习模型提取所述历史车险信息中与车辆置换相关的j维特征向量,并通过所述j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数;确定模块,用于采用所述预训练模型集合中的修正模型对所述第一概率和所述第二概率融合,得到融合概率,并根据所述融合概率,确定所述预训练模型集合对历史车辆置换的预测结果;残差处理模块,用于采用所述修正模型计算所述预测结果和历史车辆置换的真实结果之间的残差值,通过所述i维特征向量和所述残差值,对所述预测结果进行线性回归处理,得到处理结果;迭代模块,用于根据所述处理结果对所述低阶学习模型和所述高阶学习模型进行迭代,直到所述低阶学习模型和所述高阶学习模型收敛时,得到车辆置换预测模型集合;预测模块,用于获取目标车辆的车险信息,并将所述目标车辆的车险信息输入所述车 辆置换预测模型集合中,预测所述目标车辆置换的预测结果。
- 根据权利要求16所述的车辆置换预测装置,其中,所述输入模块包括:获取单元,用于获取车辆的历史车险信息;提取单元,用于当i=1时,提取所述历史车险信息中与车辆置换相关的多个特征因子和各所述特征因子对应的属性类别,并根据所述属性类别,对各所述特征因子进行分组处理,得到多个因子组合;采用预置稀疏特征向量对各所述因子组合进行编码嵌入,得到所述历史车险信息中的一维特征向量;输入单元,用于将所述i维特征向量输入预置的预训练模型集合中,其中,所述预训练模型模型集合包括一个低阶学习模型、一个高阶学习模型和一个修正模型;
- 根据权利要求16所述的车辆置换预测装置,其中,所述低阶特征融合模块包括:低阶特征融合单元,用于将所述i维特征向量作为第一基础向量,采用预置预训练模型集合中的低阶学习模型对每两个第一基础向量进行组合,得到多个向量组合;根据所述向量组合的数量,建立i维特征向量对应的交叉权重矩阵,并根据所述交叉权重矩阵,依次对各所述向量组合中的两个第一基础向量进行融合,得到对应的i+1维特征向量;第一预测单元,用于通过i+1维特征向量,预测车辆置换的第一概率,其中,i为大于等于1的正整数。
- 根据权利要求16所述的车辆置换预测装置,其中,所述高阶特征提取模块包括:高阶特征提取单元,用于将所述i维特征向量作为第二基础向量,采用所述预训练模型集合中的高阶学习模型对各所述第二基础向量进行加权组合,得到多个加权组合向量,并对各所述加权组合向量进行非线性映射处理,得到与车辆置换相关k维特征向量,其中,j>k>i;将所述k维特征向量作为新的第二基础向量,并判断所述新的第二基础向量是否满足预置跳变条件;若满足,则计算所述k维特征向量对应的残差向量,并将所述残差向量作为k+1维特征向量,将所述k+1维特征向量作为新的第二基础向量,对各所述新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止;若不满足,则对各所述新的第二基础向量进行加权组合以及非线性映射处理,直到得到j维特征向量时停止;第二预测单元,用于通过所述j维特征向量,预测车辆置换的第二概率,其中,j大于等于i的正整数。
- 根据权利要求19所述的车辆置换预测装置,其中,所述高阶特征提取单元还用于:将所述i维特征向量作为第二基础向量,并根据各所述第二基础向量的预置激活概率,采用所述预训练模型集合中的高阶学习模型随机选取多个第二基础向量;根据所述激活概率确定选取的第二基础向量对应的加权系数,并采用所述加权系数对各所述第二基础向量进行加权组合,得到多个加权组合向量。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110754151.6A CN113255842B (zh) | 2021-07-05 | 2021-07-05 | 车辆置换预测方法、装置、设备及存储介质 |
CN202110754151.6 | 2021-07-05 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023279694A1 true WO2023279694A1 (zh) | 2023-01-12 |
Family
ID=77190616
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2022/071479 WO2023279694A1 (zh) | 2021-07-05 | 2022-01-12 | 车辆置换预测方法、装置、设备及存储介质 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN113255842B (zh) |
WO (1) | WO2023279694A1 (zh) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116308828A (zh) * | 2023-05-10 | 2023-06-23 | 凯泰铭科技(北京)有限公司 | 一种出险数据集成融合的车辆智能定损方法及系统 |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113255842B (zh) * | 2021-07-05 | 2021-11-02 | 平安科技(深圳)有限公司 | 车辆置换预测方法、装置、设备及存储介质 |
CN114244611B (zh) * | 2021-12-17 | 2023-10-13 | 中国平安财产保险股份有限公司 | 异常攻击检测方法、装置、设备及存储介质 |
CN118037343B (zh) * | 2024-04-15 | 2024-07-12 | 中汽信息科技(天津)有限公司 | 车辆置换概率确定方法、设备和介质 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109784586A (zh) * | 2019-03-07 | 2019-05-21 | 上海赢科信息技术有限公司 | 车险的出险情况的预测方法及系统 |
CN109902222A (zh) * | 2018-11-30 | 2019-06-18 | 华为技术有限公司 | 一种推荐方法及装置 |
US20190219994A1 (en) * | 2018-01-18 | 2019-07-18 | General Electric Company | Feature extractions to model large-scale complex control systems |
CN113255842A (zh) * | 2021-07-05 | 2021-08-13 | 平安科技(深圳)有限公司 | 车辆置换预测方法、装置、设备及存储介质 |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11195057B2 (en) * | 2014-03-18 | 2021-12-07 | Z Advanced Computing, Inc. | System and method for extremely efficient image and pattern recognition and artificial intelligence platform |
CN110675267B (zh) * | 2018-07-03 | 2023-05-12 | 北京嘀嘀无限科技发展有限公司 | 一种根据实时路况进行车险预警的方法及系统 |
CN110569701B (zh) * | 2018-12-29 | 2020-08-07 | 阿里巴巴集团控股有限公司 | 计算机执行的车辆定损方法及装置 |
CN110033386B (zh) * | 2019-03-07 | 2020-10-02 | 阿里巴巴集团控股有限公司 | 车辆事故的鉴定方法及装置、电子设备 |
US10846716B1 (en) * | 2019-12-27 | 2020-11-24 | Capital One Services, Llc | System and method for facilitating training of a prediction model to estimate a user vehicle damage tolerance |
-
2021
- 2021-07-05 CN CN202110754151.6A patent/CN113255842B/zh active Active
-
2022
- 2022-01-12 WO PCT/CN2022/071479 patent/WO2023279694A1/zh unknown
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190219994A1 (en) * | 2018-01-18 | 2019-07-18 | General Electric Company | Feature extractions to model large-scale complex control systems |
CN109902222A (zh) * | 2018-11-30 | 2019-06-18 | 华为技术有限公司 | 一种推荐方法及装置 |
CN109784586A (zh) * | 2019-03-07 | 2019-05-21 | 上海赢科信息技术有限公司 | 车险的出险情况的预测方法及系统 |
CN113255842A (zh) * | 2021-07-05 | 2021-08-13 | 平安科技(深圳)有限公司 | 车辆置换预测方法、装置、设备及存储介质 |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116308828A (zh) * | 2023-05-10 | 2023-06-23 | 凯泰铭科技(北京)有限公司 | 一种出险数据集成融合的车辆智能定损方法及系统 |
CN116308828B (zh) * | 2023-05-10 | 2023-08-11 | 凯泰铭科技(北京)有限公司 | 一种出险数据集成融合的车辆智能定损方法及系统 |
Also Published As
Publication number | Publication date |
---|---|
CN113255842A (zh) | 2021-08-13 |
CN113255842B (zh) | 2021-11-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2023279694A1 (zh) | 车辆置换预测方法、装置、设备及存储介质 | |
Sanchez et al. | Deep learning for population size history inference: Design, comparison and combination with approximate Bayesian computation | |
Nyathi et al. | Comparison of a genetic algorithm to grammatical evolution for automated design of genetic programming classification algorithms | |
CN112070125A (zh) | 一种基于孤立森林学习的不平衡数据集的预测方法 | |
Shin et al. | A genetic algorithm application in bankruptcy prediction modeling | |
CN109948735B (zh) | 一种多标签分类方法、系统、装置及存储介质 | |
CN110674636B (zh) | 一种用电行为分析方法 | |
CN109902808B (zh) | 一种基于浮点数位变异遗传算法优化卷积神经网络的方法 | |
CN115878904A (zh) | 基于深度学习的知识产权个性化推荐方法、系统及介质 | |
US11847389B2 (en) | Device and method for optimizing an input parameter in a processing of a semiconductor | |
CN112215604A (zh) | 交易双方关系信息识别方法及装置 | |
Hsu et al. | Dynamically optimizing parameters in support vector regression: An application of electricity load forecasting | |
CN117811843B (zh) | 基于大数据分析和自主学习的网络入侵检测方法及系统 | |
CN117635238A (zh) | 一种商品推荐方法、装置、设备及存储介质 | |
Ding et al. | Dance: A deep learning library and benchmark for single-cell analysis | |
CN118037423A (zh) | 农业贷款后农户还款意愿评价方法及系统 | |
CN117093849A (zh) | 一种基于自动生成模型的数字矩阵特征分析方法 | |
CN115795035A (zh) | 基于进化神经网络的科技服务资源分类方法、系统及其计算机可读存储介质 | |
CN113571198B (zh) | 转化率预测方法、装置、设备及存储介质 | |
CN115712775A (zh) | 产品推荐方法、装置、计算机设备和存储介质 | |
CN115333957A (zh) | 基于用户行为和企业业务特征的业务流量预测方法及系统 | |
CN113590691A (zh) | 目标对象处理方法以及装置 | |
CN112884028A (zh) | 一种系统资源调整方法、装置及设备 | |
Whitehouse et al. | Tree sequences as a general-purpose tool for population genetic inference | |
CN113469450B (zh) | 一种数据分类方法、装置、计算机设备及存储介质 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 26/04/2024) |