Docs

DataWind

Common Template: Case Introduction

Prediction of Automotive User Intent Tags

1. Scenario overview

Car marketing includes a variety of intent recommendation scenarios, such as common optional package intent recommendations, potential customer car purchase intent prediction, and replacement or maintenance intent prediction. Business personnel can combine user lifecycle management to adopt different marketing strategies for user groups at different lifecycle stages, such as:

Unaware users: car brand and model introductions, driving knowledge recommendations
Users learning to drive: car brand and model introductions, understanding car purchase needs, driving knowledge, car purchase knowledge
Users planning to buy a car: car brand and model introductions, car purchase timing, car purchase budget, recommendation of intended car models, car purchase knowledge, car promotion offers
Users about to purchase a car: car brand and model introductions, car purchase timing, car purchase budget, recommendation of intended car models, car purchase knowledge, car promotion offers, lead collection
Users who have just purchased a car: car brand and model introductions, car purchase timing, brand and model, car promotional activities, lead collection, car purchase knowledge
Existing car owners: car brand and model introductions, repeat purchase or replacement needs, car purchase budget, intended car brands, replacement lead collection, car knowledge, promotional activities

This case focuses on the scenarios of users planning to purchase a car, users about to purchase a car, and existing car owners. It uses machine learning to infer users' car purchase intent, trains on historical data of car purchase and replacement users, predicts car purchase intent for users in these three lifecycle stages, and adopts appropriate marketing measures.

2. Modeling process

2.1 Data description

Field name	Field type	Field definition	Field description
uid	bigint	User ID
gender	string	Gender	Gender: Male, female
age	string	Age group	Age group: 18-23, 24-30, 31-40, 41-50, 51 and above
area	string	Permanent residence	Permanent residence - district
lifestage	string	Life stage	Life stage: Seeking marriage, preparing for marriage, pregnancy, have children
education	string	Education	Education: High school or below, associate degree, bachelor's degree, master's degree or above
job	string	Job	Job: International student, doctor, lawyer, teacher, blue-collar worker, white-collar worker, others
birthday	string	Birthday	Birthday
lifecycle	string	Lifecycle	Lifecycle: Fan, Potential customer, Car owner, Lost
own_store	string	Affiliated store	Affiliated store: Zhonghuan Guanghui, Pudong Yongda, Others
own_sales	string	Responsible Sales representative	Responsible Sales representative: Dawson, Wang Ning, Robert
wanted_store	string	Intended dealer:	Intended dealer: Pudong Yongda, Zhonghuan Guanghui
wanted_carmodel	string	Intended car model	Intended car model: RX5 Plus, ER6, RX3, i5, i6, iMAX8, Unknown
rank	string	Intention rating	Intention rating: High, Medium, Low, Completed
lasttime_pudongyongda_testdrive	string	Last store test drive time	Last store test drive time: within 1 week, within 2 weeks, within 1 month, within 2 months, within 3 months, within half a year, within one year
newcar_tasting	string	New car tasting event	New car tasting event: Yes, No
storetimes_lastthreemonth	string	Number of dealer visits in the past 3 months	Number of dealer visits in the past 3 months: 0, 1, 2, 3, 4, >4
storenumbers_lastthreemonth	string	Number of dealers visited in the past 3 months	Number of dealers visited in the past 3 months: 0, 1, 2, 3, 4, >4
warranty_times	int	Number of store visits within warranty period	Number of store visits within warranty period: 0, 1, 2, 3, 4, 5
warranty_values	string	Total service value within warranty period	Total service value within warranty period: 0-3000, 3000-5000, 5000~10000, >10000
pudongyongda_noentry_time	string	Time not entering this store	Time not entering this store (Pudong Yongda): 1 month, 2 months, 3 months, 4 months, 5 months, 6 months, 7 months, 8 months, 9 months, 10 months, 11 months, 12 months
forwardtimes_lastthreemonth	string	Total forwards in the past 3 months	Total forwards in the past 3 months: <3, 3-5, 6-10, >10
publishtimes_lastmonth	string	Number of posts in the past month	Number of posts in the past month: 0, 1, 2, 3, 4, >4
applets_opentimes_lastmonth	int	Mini program open times in the past month	Mini program open times in the past month
optionalpackpage_opentimes_lastmonth	int	Optional package page open times in the past month	Optional package page open times in the past month
optionalpackpage_time_lastmonth	int	Optional package page browsing duration in the past month	Optional package page browsing duration in the past month. Unit: min
last_eta	int	Current ETA	Current ETA. Unit: days
eta_modifytimes	string	ETA modification times	ETA modification times: 0, 1, 2, 3, 4, >4
pointuse_lastmonth	string	Used points in the past month	Used points in the past month: Yes, No
point_count	string	Points amount	Points amount: 0-500, 500-10000, 10000-30000, above 30000
active_lasttime	string	Last active time	Last active time
contact_preference	string	Contact preference	Contact preference
gift_preference	string	Preferred premium categories	Preferred premium categories
carowner	string	Owner of this vehicle	Owner of this vehicle
salescarmodel	string	Car model sold in this transaction	Car model sold in this transaction
salestime	string	Transaction date of this vehicle	Transaction date of this vehicle
carownerappvisit	string	Car owner app visit frequency	Car owner app visit frequency
othercarmodel	string	Other car models owned	Other car models owned
interest_car_price	string	Car price	Car price
interest_car_brand_country	string	Car brand country of origin	Car brand country of origin
interest_car_brand	string	Car brand	Car brand
interest_car_other_series_type	string	Car model	Car model
interest_car_newenergy_type	string	New energy type	New energy type
client_stage	string	Customer stage	Customer stage
channel	string	Lead source channel	Lead source channel
community_pv_lastweek	int	Community PV in the past week	Community PV in the past week
community_creation_lastweek	int	Community content creation in the past week	Community content creation in the past week
community_forward_lastweek	int	Community shares in the past week	Community shares in the past week
community_recruit_lastweek	int	Community new user acquisition in the past week	Community new user acquisition in the past week
car_optical_pack_eagerness	string	Optional package preference	Optional package preference
car_pick_up_eagerness	string	Car pick-up urgency	Car pick-up urgency
returned_customer	string	Returning customer	Returning customer
car_price_comparison	string	Cross-store price comparison	Cross-store price comparison
stored_many_points	string	Large accumulation of points	Large accumulation of points
club_freshman	string	New club member	New club member
recall_values	string	Churn recall value	Churn recall value

The data contains most of the key processed indicators, including user lifecycle, activity level, channel information, personal attributes, and more. By leveraging such structured data and applying straightforward modeling, it is possible to predict intent ratings and thus enable precise marketing.

2.2 Data preparation

Data access is achieved through a direct connection to Hive data. This case involves two datasets: historical manually labeled data and newly generated data. Classification training is performed on the historical labeled data for intention rating, and intention prediction is carried out on the new data.

2.3 Problem modeling

Core indicators are selected for intention training. Since data processing has already been completed in advance, all indicators can be considered as core factors influencing the selection of intention. Therefore, there is no need for feature selection or feature processing here. The target requirements can be met through simple logistic regression.

2.4 Model evaluation

Supports model evaluation during training, assisting users with manual parameter tuning. Refer to the confusion matrix below, which shows the final recall and precision of the model, as well as the performance on the test set under different classification scenarios. It also supports the calculation of weighted recall, weighted precision, and accuracy.

2.5 Data production application

After obtaining a model with satisfactory performance, the prediction results can be exported to Hive/ClickHouse. Subsequent visualization queries and dashboard creation can be performed on the DataWind platform. It can also be integrated with CDP products for tag insights, marketing applications, user segmentation, and more.

Last updated: 2026.06.09 11:33:36

DataWind

2.1 Data description #

2.2 Data preparation #

2.3 Problem modeling #

2.4 Model evaluation #

2.5 Data production application #

2.1 Data description

2.2 Data preparation

2.3 Problem modeling

2.4 Model evaluation

2.5 Data production application