Methods, devices, storage media, and equipment for predicting purchase intentions for new energy vehicles
By combining pre-trained classification models and multiple models, and using proprietary and full-scale data to predict car purchase intentions, the problem of low accuracy and high cost in predicting new energy vehicle purchase intentions has been solved, achieving efficient and accurate car purchase intention prediction.
Patent Information
- Application Number
- CN202611452864.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-18
- Publication Date
- 2026-11-10
AI Technical Summary
The existing prediction of purchase intentions for new energy vehicles is inaccurate and the cost of data procurement is high. Existing user profiling technologies also suffer from problems such as one-sided data sources or single feature dimensions.
By acquiring customer business data, using a pre-trained classification model for preliminary prediction, determining the target value range, and then retrieving external data within this range for full prediction, the purchase intention prediction result is obtained by weighted fusion of proprietary and full prediction models.
It improves the accuracy of car purchase intention prediction, reduces data procurement and interface call costs, simplifies the prediction process, adapts to high-concurrency scenarios, and meets the diverse car purchase intention needs of new energy vehicles.
Smart Images

Figure CN123071010A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of financial technology, and in particular to a method, apparatus, storage medium, and device for predicting purchase intentions for new energy vehicles. Background Technology
[0002] With the deepening integration of internet technology into the new energy vehicle business, online marketing has become a key channel for new energy vehicle marketing. In particular, financial institutions such as banks urgently need to accurately identify customers with purchasing intentions when conducting online promotions, thereby achieving precise delivery of marketing information and effectively avoiding resource waste caused by misdirection.
[0003] Currently, the prediction of existing customers' car purchase intentions is typically based on user profiling technology. However, user profiling data sources are often incomplete or lack comprehensive feature dimensions, failing to uncover full and multi-dimensional user information. This results in low accuracy in predicting customer purchase intentions, and adding too many profiling conditions increases data acquisition costs. Therefore, there is an urgent need for a method to predict the purchase intentions of new energy vehicles to address these issues. Summary of the Invention
[0004] In view of this, this application provides a method, device, storage medium and equipment for predicting the purchase intention of new energy vehicles, the main purpose of which is to solve the problem of poor accuracy of existing purchase intention predictions.
[0005] According to one aspect of this application, a method for predicting purchase intentions for new energy vehicles is provided, comprising: Obtain customer business data; The business data is classified and predicted based on a pre-trained classification model, and the classification prediction results are obtained: When the classification prediction result is within the target value range, the customer's external data is retrieved, and the business data and the external data are predicted based on the full prediction model that has completed model training to obtain the car purchase intention prediction result. The target value range is determined based on the gain threshold corresponding to the classification model.
[0006] Furthermore, before obtaining the classification prediction result by classifying the business data using the pre-trained classification model, the method further includes: Obtain training samples for vehicle insurance business, the training samples including proprietary samples and full samples, the full samples including proprietary samples and external samples; Based on the proprietary samples and the full sample, the multi-model group is trained to obtain the proprietary prediction model and the full prediction model. The proprietary prediction model and the full prediction model are then used to predict the training samples to obtain the proprietary prediction probability and the full prediction probability. The classification model is trained using the difference between the proprietary prediction probability and the full prediction probability as training labels, and the gain threshold is determined based on the prediction results output by the classification model on the training samples. The target value range is determined based on the gain threshold.
[0007] Further, the step of predicting the training samples using the proprietary prediction model and the full prediction model to obtain the proprietary prediction probability and the full prediction probability includes: Based on the proprietary prediction model that has completed model training, a proprietary sample prediction is performed to obtain a first prediction result, and based on the full sample prediction model that has completed model training, a full sample prediction is performed to obtain a second prediction result. The first prediction result and the second prediction result are weighted and fused respectively to obtain the fused independent prediction probability and the full prediction probability.
[0008] Further, determining the target value range interval based on the gain threshold includes: Determine the sample objects of the training samples; The gain prediction result of the sample object is compared with the value of the gain threshold, and the target value range is determined based on the comparison result. The target value range includes the maximum value of the range and the minimum value of the range.
[0009] Furthermore, the method also includes: The model accuracy for classifying different sample objects based on the classification model is obtained according to the gain threshold. The value of the gain threshold corresponding to the sample object is determined based on the gain threshold corresponding to the maximum model accuracy.
[0010] Furthermore, after the pre-trained classification model performs classification prediction on the business data to obtain the classification prediction result, the method further includes: Determine the target value range corresponding to the business data; If the business data includes multiple data items, then compare the classification prediction results corresponding to each of the multiple data items one by one to see if they are within the corresponding target threshold range. When any of the classification prediction results is not within the target value range, the business data is predicted based on the proprietary prediction model that has completed model training, and the car purchase intention prediction result is obtained.
[0011] Furthermore, the multi-model group includes the Xgboost model, LightGBM model, CatBoost model, random forest model, and logistic regression model.
[0012] According to another aspect of this application, a device for predicting the purchase intention of new energy vehicles is provided, comprising: The acquisition module is used to acquire customer business data; The classification module is used to perform classification prediction on the business data based on a pre-trained classification model, and obtain the classification prediction result: The prediction module is used to retrieve the customer's external data when the classification prediction result is within the target value range, and to predict the business data and the external data based on the full prediction model that has been trained, so as to obtain the car purchase intention prediction result. The target value range is determined based on the gain threshold corresponding to the classification model.
[0013] Furthermore, the device also includes: a training module and a determination module. The acquisition module is also used to acquire training samples of vehicle insurance business. The training samples include proprietary samples and full samples. The full samples include proprietary samples and external samples. The training module is used to train multiple model groups based on the proprietary samples and the full samples respectively to obtain proprietary prediction models and full prediction models, and to predict the training samples using proprietary prediction models and full prediction models to obtain proprietary prediction probabilities and full prediction probabilities. The determining module is used to train the classification model using the proprietary prediction probability and the full prediction probability as training labels, and to determine a gain threshold based on the prediction results output by the classification model on the training samples; and to determine the target value range based on the gain threshold.
[0014] Furthermore, the training module is specifically used to perform proprietary sample prediction based on the proprietary prediction model that has completed model training, to obtain a first prediction result, and to perform full sample prediction based on the full prediction model that has completed model training, to obtain a second prediction result; and to perform weighted fusion of the first prediction result and the second prediction result respectively, to obtain the fused proprietary prediction probability and the full prediction probability.
[0015] Furthermore, the determining module is specifically used to determine the sample object of the training sample; compare the gain prediction result of the sample object with the value result of the gain threshold, and determine the target value range based on the comparison result, wherein the target value range includes the maximum value of the range and the minimum value of the range.
[0016] Furthermore, the determining module is specifically used to obtain the model accuracy of different sample objects classified based on the classification model according to the gain threshold; and to determine the value of the gain threshold corresponding to the sample object based on the gain threshold corresponding to the maximum model accuracy.
[0017] Furthermore, the device also includes: The comparison module is used to determine the target value range corresponding to the business data; if the business data includes multiple data, the classification prediction results corresponding to the multiple data are compared one by one to see if they are within the corresponding target threshold range; if any of the classification prediction results are not within the target value range, the business data is predicted based on the proprietary prediction model that has completed model training to obtain the car purchase intention prediction result.
[0018] Furthermore, the multi-model group includes the Xgboost model, LightGBM model, CatBoost model, random forest model, and logistic regression model.
[0019] According to another aspect of this application, a storage medium is provided, wherein at least one executable instruction is stored therein, the executable instruction causing a processor to perform an operation corresponding to the above-described method for predicting the purchase intention of new energy vehicles.
[0020] According to another aspect of this application, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-mentioned method for predicting the purchase intention of new energy vehicles.
[0021] By employing the above technical solutions, the technical solutions provided in the embodiments of this application have at least the following advantages: This application provides a method, apparatus, storage medium, and device for predicting purchase intentions for new energy vehicles. Compared with the prior art, the embodiments of this application obtain customer business data; classify and predict the business data based on a pre-trained classification model to obtain a classification prediction result. The classification model is used to determine the credibility of the business data. When the classification prediction result is within a target value range, external data of the customer is retrieved, and the business data and the external data are predicted based on a fully trained prediction model to obtain a purchase intention prediction result. The target value range is determined based on the gain threshold corresponding to the classification model. By intelligently judging the prediction gain of external data introduced into the business data through the pre-trained classification model, the cost of data procurement and interface calls is effectively reduced, the prediction process is simplified, the response latency is shortened, and it is adapted to high-concurrency car insurance purchase intention prediction scenarios, thereby meeting the diverse purchase intention prediction needs of new energy vehicles.
[0022] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description
[0023] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a method for predicting purchase intentions for new energy vehicles, as provided in an embodiment of this application, is shown. Figure 2 This illustration shows a schematic diagram of a classification model training and prediction process provided in an embodiment of this application; Figure 3 This paper shows a block diagram of a device for predicting the purchase intention of a new energy vehicle according to an embodiment of this application. Figure 4 A schematic diagram of the structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation
[0024] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0026] The embodiments of this invention can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.
[0027] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.
[0028] Based on this, in one embodiment, the present invention provides a method for predicting the purchase intention of new energy vehicles. Taking the application of this method to computer equipment such as servers as an example, the server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.
[0029] This application provides a method for predicting purchase intentions for new energy vehicles, such as... Figure 1 As shown, the method includes: 101. Obtain customer business data.
[0030] In this embodiment, the current execution entity serves as the execution end for predicting car purchase intentions. It can connect with the business system or be a prediction module embedded in the business system to obtain customer business data. In this case, business data refers to variables related to new energy vehicle insurance, such as current vehicle information, historical vehicle information, and other commercial insurance coverage. This embodiment does not impose specific limitations.
[0031] In one specific embodiment, current vehicle information includes vehicle price, vehicle age, brand, number of vehicles registered under one's name, mileage, and accident history. Historical vehicle information includes vehicle price, ownership age, historical number of vehicles, and historical accident history. Other commercial insurance information may include the number of recent personal property and life insurance policies, premium payments, and insurance claims, as well as the number of recent property and life insurance policies, premium payments, insurance claims, and the qualifications of the employing company. The above business data can be directly retrieved from the business system or entered directly by the customer. This embodiment does not impose specific limitations.
[0032] 102. Based on the pre-trained classification model, classify and predict the business data to obtain the classification prediction results.
[0033] In this embodiment, the classification model is used to determine whether external data needs to be called when predicting intentions based on business data. That is, the classification model is used to classify and predict business data to obtain classification prediction results. At this time, the classification prediction results are numerical probability values, representing the estimated results of the gain introduced by external data. In addition, the classification model can be a pre-trained classifier model such as a random forest model or a logistic regression model, or it can be the same model structure as the Xgboost model, LightGBM model, CatBoost model, random forest model, and logistic regression model included in the proprietary prediction model and the full prediction model. In this case, the proprietary prediction model and the full prediction model are used to predict car purchase intentions. When the classification model, the proprietary prediction model, and the full prediction model all use five models from the multi-model group, the training labels used are different, thereby achieving different prediction purposes. This embodiment does not make specific limitations.
[0034] In another embodiment of this application, for further definition and explanation, before the step of classifying and predicting the business data based on a pre-trained classification model to obtain the classification prediction result, the method further includes: Obtain training samples for vehicle insurance business; Based on the proprietary samples and the full sample, the multi-model group is trained to obtain the proprietary prediction model and the full prediction model. The proprietary prediction model and the full prediction model are then used to predict the training samples to obtain the proprietary prediction probability and the full prediction probability. The classification model is trained using the difference between the proprietary prediction probability and the full prediction probability as training labels, and the gain threshold is determined based on the prediction results output by the classification model on the training samples. The target value range is determined based on the gain threshold.
[0035] To construct a target value range for judging the availability of business data and thereby improve the accuracy of predicting car purchase intentions using business data, the current execution terminal pre-acquires training samples of vehicle insurance business. This can be done based on business data cached in the vehicle insurance business system; this embodiment does not specify a particular method. The training samples include proprietary samples and the full set of samples. The full set of samples includes proprietary samples and external samples. Then, multiple model groups are trained based on the proprietary samples and the full set of samples. These multiple model groups include XgBoost, LightGBM, CatBoost, Random Forest, and Logistic Regression models.
[0036] In this embodiment of the application, since the multi-model group contains multiple models with classification functions, after model training, a set of multi-model groups that have completed training is obtained for the proprietary samples, and a set of multi-model groups that have completed training is obtained for the full set of samples, namely the proprietary prediction model and the full set of prediction models, so as to make predictions based on the proprietary prediction model and the full set of prediction models respectively, and obtain the proprietary prediction probability and the full set of prediction probability.
[0037] Furthermore, such as Figure 2 As shown, the trained proprietary prediction model and the full prediction model are used to predict the training samples, respectively, to obtain the proprietary prediction probability and the full prediction probability for each training sample. The probability difference between the two is calculated, which is the difference between the full prediction probability and the proprietary prediction probability. The classification model is trained using the proprietary samples based on the probability difference as the training label. At this time, the samples can be divided separately for prediction and training to distinguish between the training set and the prediction set. This embodiment of the application does not make specific limitations.
[0038] In a specific training embodiment, the classification model is trained using proprietary business data of its own samples as input variables and probability differences as the prediction target. The classification model also employs a multi-model soft-voting fusion mode, using five base learners: XGBoost, LightGBM, CatBoost, Random Forest, and Linear Regression. Each base learner is trained as an independent sub-model with its own data variables as input. Finally, the outputs of the five sub-models are weighted and soft-voted to obtain the classification model, whose output is a predicted value of the external data gain for a single sample.
[0039] After training the classification model, the model is used to predict the gain of each sample in the training set, and the gain threshold is determined. Specifically, the gain estimates of all training samples can be sorted in ascending order, and the gain threshold P can be set to the 90th quantile. Based on this, the target value range is determined to determine whether the sample falls within the target value range.
[0040] In another embodiment of this application, for further definition and explanation, the step of predicting the training samples using the proprietary prediction model and the full prediction model to obtain the proprietary prediction probability and the full prediction probability includes: Based on the proprietary prediction model that has completed model training, a proprietary sample prediction is performed to obtain a first prediction result, and based on the full sample prediction model that has completed model training, a full sample prediction is performed to obtain a second prediction result. The first prediction result and the second prediction result are weighted and fused respectively to obtain the fused independent prediction probability and the full prediction probability.
[0041] To improve the accuracy and stability of the proprietary prediction probability and the full prediction probability, and to provide reliable data support for subsequent calculations of the difference between the two prediction probabilities, construction of classification model training labels, and determination of gain thresholds, the current execution end, when predicting training samples using both the proprietary prediction model and the full prediction model, specifically, firstly, performs predictions based on the proprietary prediction model that has completed model training. Since the multi-model group contains five base learners—XGBoost, LightGBM, CatBoost, Random Forest, and Linear Regression—that operate independently and in parallel, each base learner outputs an independent initial prediction probability for the same proprietary sample based on pre-converged model parameters. All initial prediction probabilities are used as the first prediction result for the proprietary sample. Simultaneously, based on the full prediction model that has completed model training, the five uniformly configured base learners are reused for independent parallel predictions. Each base learner outputs a corresponding initial prediction probability for the full sample features, and the prediction results of all base learners are used as the second prediction result for the full sample.
[0042] In this embodiment, after obtaining the first and second prediction results, a weighted fusion process is performed on the two sets of prediction results, which can be represented as a soft voting method. For the first prediction result, weight normalization is performed based on the comprehensive prediction performance of each base learner on the validation set to ensure that the sum of the fusion weights of all base learners is 1. The initial prediction probabilities of the five types of base learners are fused and calculated using a weighted summation method, finally obtaining the fused individual prediction probabilities. At this point, the output prediction probability is expressed as... The fusion weight is Finally, based on the least squares solution method, the inherent prediction probability formula is solved, which is expressed as: ; in, The true probability is calculated as follows. For the second prediction result, a weighted fusion strategy and weighting rules consistent with the prediction results of the own samples are adopted to perform weighted fusion on the initial prediction probabilities of the full sample output by XGBoost, LightGBM, CatBoost, Random Forest, and Linear Regression. This avoids the prediction limitations and random biases of a single algorithm model and obtains the fused full prediction probability. The embodiments of this application will not be described in detail here.
[0043] In another embodiment of this application, for further definition and explanation, the step of determining the target value range based on the gain threshold includes: Determine the sample objects of the training samples; The gain prediction result of the sample object is compared with the value of the gain threshold, and the target value range is determined based on the comparison result.
[0044] To improve the accuracy of car purchase intention prediction by using a target threshold range as the basis for distinguishing between proprietary and external data, the current execution end first determines the sample objects for the training samples participating in the threshold range calibration when determining the target value range. These sample objects consist of all historical customer samples related to vehicle insurance business. Each sample object is pre-matched with corresponding proprietary business feature data, gain prediction results, and model-estimated gain values, forming a complete threshold calibration sample set. Then, based on the obtained gain threshold, the gain threshold for each sample object is determined, resulting in the corresponding value. Specifically, the gain prediction results of each sample object X are processed by the classification model. The results are compared one by one with the calculated gain threshold, and the gain threshold is used as the critical judgment criterion. That is, the gain prediction result of each sample object X output by the classification model is compared with the value of the gain threshold. Each sample is compared with a gain threshold value, and samples whose gain prediction results are greater than the gain threshold are selected as a subset. The target value range is determined by this subset. Specifically, the target value range for each sample object is the actual value range of each sample object in that subset, i.e., the target value range can be expressed as: ; Where q is the gain threshold. This represents the result of the gain threshold selection, with a sample size of I and a number of sample variables of J. In a specific scenario, for new business data... Each has j variables, when or When it is null, record This indicates that the classification prediction result falls within this range; otherwise, it is... .
[0045] In another embodiment of this application, for further definition and explanation, the steps also include: The model accuracy for classifying different sample objects based on the classification model is obtained according to the gain threshold. The value of the gain threshold corresponding to the sample object is determined based on the gain threshold corresponding to the maximum model accuracy.
[0046] In this embodiment, after the classification model is trained, the optimal gain threshold is selected and determined based on the trained classification model, which improves the effectiveness of determining the optimal gain threshold. Specifically, multiple preset gain threshold values q can be iterated. For each threshold q to be tested, all training sample objects can be input into the trained classification model to obtain the gain prediction result corresponding to each sample object. Based on the current threshold q, the gain of all sample objects is classified into high and low categories. The overall classification accuracy of the model corresponding to each set of thresholds q is statistically analyzed to construct a dataset showing the correspondence between thresholds and model accuracy. For example, a line graph showing the relationship between thresholds and accuracy can be plotted to visually represent the fluctuation pattern of model accuracy as the gain threshold q changes.
[0047] In one specific embodiment, for the model accuracy line graphs corresponding to multiple sets of thresholds, the threshold intervals in which the accuracy has not decreased significantly can be screened, and the largest q value within this effective interval can be selected. This q value is determined as the optimal gain threshold, and thus used as the final value of the sample object gain threshold.
[0048] In a test example, comparative verification showed that setting the gain threshold q to 0.99 was the optimal value. This value represents the maximum threshold value without a significant decrease in accuracy, fully meeting the model's prediction accuracy constraints. For instance, when the gain threshold q is 0.99, the AUC of both the classification model and the overall prediction model decreases only slightly by one to two-thousandths of a point, while the prediction accuracy remains essentially unchanged. Simultaneously, it reduces the frequency of external data calls by 22% to 24%, significantly lowering the interface overhead, traffic costs, and business data usage costs associated with external data retrieval, while ensuring model stability.
[0049] 103. When the classification prediction result is within the target value range, retrieve the customer's external data, and predict the business data and the external data based on the full prediction model that has completed model training, to obtain the car purchase intention prediction result.
[0050] In this embodiment, after the current execution end obtains the classification prediction result, it determines whether the classification prediction result is within the target value range. The target value range is determined based on the gain threshold and different training samples; that is, during training, it is preferably determined based on the threshold range corresponding to the gain prediction value. When the classification prediction result is not within the target value range, it indicates that the current business data can be directly used for purchase intention prediction. Conversely, when the classification prediction result is within the target value range, it indicates that the business data in the proprietary business system cannot meet the data requirements for direct purchase intention prediction. Therefore, external customer data is retrieved, and a full-scale prediction model based on the completed model training is used to predict both the business data and the external data to obtain the purchase intention prediction result.
[0051] It should be noted that the external data in this application embodiment refers to data that can be obtained outside of the proprietary business system, including but not limited to personal income, credit history, debt risk, consumption preferences, payment characteristics and other data closely related to personal large-scale consumption. In this case, an external data acquisition request can be initiated to request data from an external system. This application embodiment does not make specific limitations.
[0052] The full prediction model in this application refers to the car purchase intention prediction model obtained after model training based on external samples and own samples. This application does not make specific limitations.
[0053] In another embodiment of this application, for further definition and explanation, after the step of classifying and predicting the business data based on a pre-trained classification model and obtaining the classification prediction result, the method further includes: Determine the target value range corresponding to the business data; If the business data includes multiple data items, then compare the classification prediction results corresponding to each of the multiple data items one by one to see if they are within the corresponding target threshold range. When any of the classification prediction results is not within the target value range, the business data is predicted based on the proprietary prediction model that has completed model training, and the car purchase intention prediction result is obtained.
[0054] In one embodiment of this application, after obtaining the classification prediction result, to identify whether external data needs to be introduced, for the customer vehicle insurance business data to be predicted, if a single business data item contains multi-dimensional proprietary feature data, it can be judged independently for each dimension. That is, the business data of each dimension is input into the pre-trained classification model, and the classification prediction result corresponding to each dimension is output one by one, that is, the external data estimated gain value corresponding to each dimension feature, and each classification prediction result is checked in turn to see if it falls within the preset target value range. At this time, if any classification prediction result is not within the target value range, it is determined that the introduction of external data for this customer sample as a whole does not have effective prediction gain, and there is no need to start the external data retrieval process. This avoids the misjudgment problem caused by global average judgment, accurately intercepts the external data call behavior of low-gain and no-gain samples, and effectively controls the data call cost and system computing overhead while ensuring the stability of the car purchase intention prediction accuracy, adapting to the online high-efficiency prediction scenario of multi-dimensional business data.
[0055] In one embodiment of this application, the business data to be predicted for a customer typically includes multi-dimensional data. For business samples containing multiple data items, an item-by-item independent discrimination method can be adopted. That is, the classification prediction results output by the pre-trained classification model for each data item are obtained one by one, i.e., the external data gain prediction results corresponding to each data item are obtained. Then, the gain prediction results corresponding to each data item are compared to see if they are within the preset target value range, thus completing the gain validity determination of the single-dimensional data. At this time, in the process of interval verification of the classification prediction results of multiple data items one by one, if any classification prediction result is detected not to fall into the target value range, it is determined that the model prediction gain after introducing external data for the current customer sample is limited, and calling external data cannot effectively improve the prediction accuracy, but only increases the business data cost and interface call overhead. At this time, there is no need to call the customer's external data. All the business data of the current customer are directly input into the proprietary prediction model that has been trained offline, so as to complete the car purchase intention prediction calculation based on the proprietary prediction model, and finally output a stable and effective car purchase intention prediction result.
[0056] This application provides a method for predicting purchase intentions for new energy vehicles. Compared with existing technologies, this application obtains customer business data; classifies and predicts the business data based on a pre-trained classification model to obtain a classification prediction result, whereby the classification model is used to determine the credibility of the business data; when the classification prediction result is within a target value range, it retrieves the customer's external data and predicts both the business data and the external data based on a fully trained prediction model to obtain a purchase intention prediction result. The target value range is determined based on the gain threshold corresponding to the classification model. By intelligently judging the prediction gain of external data introduced from the business data through the pre-trained classification model, the method effectively reduces data procurement and interface call costs, simplifies the prediction process, shortens response latency, and adapts to high-concurrency purchase intention prediction scenarios, thereby meeting the diverse purchase intention prediction needs of new energy vehicles.
[0057] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this application provides a device for predicting the purchase intention of new energy vehicles, such as... Figure 3 As shown, the device includes: Module 21 is used to acquire customer business data; Classification module 22 is used to perform classification prediction on the business data based on a pre-trained classification model to obtain the classification prediction result: The prediction module 23 is used to retrieve the customer's external data when the classification prediction result is within the target value range, and to predict the business data and the external data based on the full prediction model that has been trained, so as to obtain the car purchase intention prediction result. The target value range is determined based on the gain threshold corresponding to the classification model.
[0058] Furthermore, the device also includes: a training module and a determination module. The acquisition module is also used to acquire training samples of vehicle insurance business. The training samples include proprietary samples and full samples. The full samples include proprietary samples and external samples. The training module is used to train multiple model groups based on the proprietary samples and the full samples respectively to obtain proprietary prediction models and full prediction models, and to predict the training samples using proprietary prediction models and full prediction models to obtain proprietary prediction probabilities and full prediction probabilities. The determining module is used to train the classification model using the proprietary prediction probability and the full prediction probability as training labels, and to determine a gain threshold based on the prediction results output by the classification model on the training samples; and to determine the target value range based on the gain threshold.
[0059] Furthermore, the training module is specifically used to perform proprietary sample prediction based on the proprietary prediction model that has completed model training, to obtain a first prediction result, and to perform full sample prediction based on the full prediction model that has completed model training, to obtain a second prediction result; and to perform weighted fusion of the first prediction result and the second prediction result respectively, to obtain the fused proprietary prediction probability and the full prediction probability.
[0060] Furthermore, the determining module is specifically used to determine the sample object of the training sample; compare the gain prediction result of the sample object with the value result of the gain threshold, and determine the target value range based on the comparison result, wherein the target value range includes the maximum value of the range and the minimum value of the range.
[0061] Furthermore, the determining module is specifically used to obtain the model accuracy of different sample objects classified based on the classification model according to the gain threshold; and to determine the value of the gain threshold corresponding to the sample object based on the gain threshold corresponding to the maximum model accuracy.
[0062] Furthermore, the device also includes: The comparison module is used to determine the target value range corresponding to the business data; if the business data includes multiple data, the classification prediction results corresponding to the multiple data are compared one by one to see if they are within the corresponding target threshold range; if any of the classification prediction results are not within the target value range, the business data is predicted based on the proprietary prediction model that has completed model training to obtain the car purchase intention prediction result.
[0063] Furthermore, the multi-model group includes the Xgboost model, LightGBM model, CatBoost model, random forest model, and logistic regression model.
[0064] This application provides a device for predicting the purchase intention of new energy vehicles. Compared with the prior art, this application obtains customer business data; classifies and predicts the business data based on a pre-trained classification model to obtain a classification prediction result. The classification model is used to determine the credibility of the business data. When the classification prediction result is within a target value range, external data of the customer is retrieved, and the business data and the external data are predicted based on a fully trained prediction model to obtain a purchase intention prediction result. The target value range is determined based on the gain threshold corresponding to the classification model. By intelligently judging the prediction gain of external data introduced into the business data through the pre-trained classification model, the cost of data procurement and interface calls is effectively reduced, the prediction process is simplified, the response latency is shortened, and it is adapted to high-concurrency car insurance purchase intention prediction scenarios, thereby meeting the diverse purchase intention prediction needs of new energy vehicles.
[0065] According to one embodiment of this application, a storage medium is provided, the storage medium storing at least one executable instruction, which can execute the method for predicting the purchase intention of new energy vehicles in any of the above method embodiments.
[0066] Figure 4 The diagram shows a structural schematic of a computer device according to one embodiment of the present application. The specific embodiments of the present application do not limit the specific implementation of the computer device.
[0067] like Figure 4 As shown, the computer device may include: a processor 302, a communications interface 304, a memory 306, and a communications bus 308.
[0068] The processor 302, communication interface 304, and memory 306 communicate with each other via communication bus 308.
[0069] Communication interface 304 is used to communicate with other network elements such as clients or other servers.
[0070] The processor 302 is used to execute program 310, which can specifically execute the relevant steps in the above-described embodiment of the method for predicting the purchase intention of new energy vehicles.
[0071] Specifically, program 310 may include program code that includes computer operation instructions.
[0072] Processor 302 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.
[0073] Memory 306 is used to store program 310. Memory 306 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.
[0074] Specifically, program 310 can be used to cause processor 302 to perform the following operations: Obtain customer business data; The business data is classified and predicted based on a pre-trained classification model, and the classification prediction results are obtained: When the classification prediction result is within the target value range, the customer's external data is retrieved, and the business data and the external data are predicted based on the full prediction model that has completed model training to obtain the car purchase intention prediction result. The target value range is determined based on the gain threshold corresponding to the classification model.
[0075] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.
[0076] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for predicting purchase intentions for new energy vehicles, characterized in that, include: Obtain customer business data; The business data is classified and predicted based on a pre-trained classification model to obtain the classification prediction result. The classification model is trained using the difference between the prediction probability of the full prediction model and the prediction probability of the proprietary prediction model as training labels. When the classification prediction result is within the target value range, the customer's external data is retrieved, and the business data and the external data are predicted based on the full prediction model that has completed model training to obtain the car purchase intention prediction result. The target value range is determined based on the gain threshold corresponding to the classification model.
2. The method according to claim 1, characterized in that, Before obtaining the classification prediction result by classifying and predicting the business data using the pre-trained classification model, the method further includes: Obtain training samples for vehicle insurance business, the training samples including proprietary samples and full samples, the full samples including proprietary samples and external samples; Based on the proprietary samples and the full samples, the multi-model group is trained to obtain the proprietary prediction model and the full prediction model. The proprietary prediction model and the full prediction model are then used to predict the training samples to obtain the proprietary prediction probability and the full prediction probability. The classification model is trained using the difference between the proprietary prediction probability and the full prediction probability as training labels, and the gain threshold is determined based on the prediction results output by the classification model on the training samples. The target value range is determined based on the gain threshold.
3. The method according to claim 2, characterized in that, The step of predicting the training samples using the proprietary prediction model and the full prediction model to obtain the proprietary prediction probability and the full prediction probability includes: Based on the proprietary prediction model that has completed model training, a proprietary sample prediction is performed to obtain a first prediction result, and based on the full sample prediction model that has completed model training, a full sample prediction is performed to obtain a second prediction result. The first prediction result and the second prediction result are weighted and fused respectively to obtain the fused independent prediction probability and the full prediction probability.
4. The method according to claim 3, characterized in that, Determining the target value range based on the gain threshold includes: Determine the sample objects of the training samples; The gain prediction result of the sample object is compared with the value of the gain threshold, and the target value range is determined based on the comparison result. The target value range includes the maximum value of the range and the minimum value of the range.
5. The method according to claim 4, characterized in that, The method further includes: The model accuracy for classifying different sample objects based on the classification model is obtained according to the gain threshold. The value of the gain threshold corresponding to the sample object is determined based on the gain threshold corresponding to the maximum model accuracy.
6. The method according to claim 1, characterized in that, After the pre-trained classification model performs classification prediction on the business data and obtains the classification prediction result, the method further includes: Determine the target value range corresponding to the business data; If the business data includes multiple data items, then compare the classification prediction results corresponding to each of the multiple data items one by one to see if they are within the corresponding target threshold range. When any of the classification prediction results is not within the target value range, the business data is predicted based on the proprietary prediction model that has completed model training, and the car purchase intention prediction result is obtained.
7. The method according to claim 1, characterized in that, The multi-model group includes the Xgboost model, LightGBM model, CatBoost model, random forest model, and logistic regression model.
8. A device for predicting purchase intentions for new energy vehicles, characterized in that, include: The acquisition module is used to acquire customer business data; The classification module is used to perform classification prediction on the business data based on a pre-trained classification model, and obtain the classification prediction result: The prediction module is used to retrieve the customer's external data when the classification prediction result is within the target value range, and to predict the business data and the external data based on the full prediction model that has been trained, so as to obtain the car purchase intention prediction result. The target value range is determined based on the gain threshold corresponding to the classification model.
9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 1.
10. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method of claim 1.
