CN117474581A

CN117474581A - Method and system for acquiring intention clients based on combat defeat clues of automobile industry

Info

Publication number: CN117474581A
Application number: CN202311384213.4A
Authority: CN
Inventors: 韩婷; 吴凡; 罗晴月; 文跃; 黄黎
Original assignee: Chongqing Changan Automobile Co Ltd
Current assignee: Chongqing Changan Automobile Co Ltd
Priority date: 2023-10-24
Filing date: 2023-10-24
Publication date: 2024-01-30

Abstract

The invention relates to a method and a system for acquiring intent clients based on combat defeat clues in the automobile industry, which are used for screening in-mould conditions, and determining clients with intent to stay according to the in-mould conditions, wherein the in-mould conditions are clients with full-network combat defeat and no actions of stay, store arrival, test driving, order placing and success; determining the probability of having the behavioral intention of arriving at a store, testing driving, placing orders and making transactions according to the user characteristics of the customer with the reserved intention; and determining whether the client has the intention of purchasing the vehicle conversion according to the probability. The invention screens out customers with the reserve intention by re-mining and utilizing the existing defeated clue resources, analyzes the probability of having the intention of going to store, trial driving, placing orders and getting in and out according to the user characteristics of the customers, and further determines whether the customers have the intention of purchasing vehicles, so that the sales strategy can be put on targeted target customers, and data support is provided for the sales terminal to assign the sales strategy.

Description

Method and system for acquiring intention clients based on combat defeat clues of automobile industry

Technical Field

The invention relates to the technical field of big data, in particular to a method and a system for acquiring an intention client based on a combat defeat clue in the automobile industry.

Background

The defeat clue refers to a sales clue marked by the dealer as no intention to purchase the vehicle or as purchased the vehicle when first followed, but in fact there is a large number of false defeats among them, i.e. customers who may also create sales opportunities, which refer to different degrees of intention for the customer to go to a store/test drive/order/deal. The cause of the false combat of the clients is various, and the willingness of the clients also gradually changes along with external factors. By using the machine learning model and the deep learning model, partial potential fake fighter clients can be mined. The defeat clue activation model is applied to the sales clue management and exchange field, and helps enterprises to better manage and utilize sales clue resources.

It should be noted that the battle cue activation model does not completely avoid the risk of cue battle, but can effectively increase the utilization rate and opportunity for success of sales cues, thereby increasing sales performance. Therefore, in practical application, other sales intelligence technologies and strategies are required to be combined to maximally utilize the existing sales lead resources, so as to improve sales performance.

The prior art provides a model generation method, an information generation method, an electronic device and a storage medium, wherein the method comprises the following steps: acquiring vehicle comment data, and extracting defeat relation information from the vehicle comment data, wherein the defeat relation information comprises main bid product models, other bid product models and attention factors of the vehicle comment data; labeling the vehicle comment data based on the combat relationship information to obtain labeling result data, and storing the vehicle comment data and the labeling result data as training samples into a training sample set; and taking the vehicle comment data of the training samples in the training sample set as input, taking the labeling result data corresponding to the input vehicle comment data as expected output, and training to obtain a combat relationship analysis model.

The prior art mainly provides a method for extracting defeat relation information from vehicle comment data, inputting target vehicle comment data into a defeat relation analysis model to obtain defeat relation information aiming at the target vehicle comment data, but the vehicle comment data has personal subjectivity and biased views, the comment data has low matching rate with clue clients, abnormal information is mixed in comment texts to cause poor data quality, and further, the comment data is classified and output in the form of taking marking result data corresponding to the input vehicle comment data as expected output, and therefore, customers cannot be distinguished from actual vehicle purchasing will according to the defeat information, so that the utilization rate of the defeat information is low, and more effective sales strategies cannot be specified by combining the output information.

Disclosure of Invention

The invention aims to provide an intention client acquisition method based on a combat defeat clue in the automobile industry, so as to solve the problems that the prior art has low utilization rate of combat defeat information and cannot be used for a sales terminal.

In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:

a method for acquiring an intention client based on a combat defeat clue in the automobile industry,

screening a model entering condition, and determining clients with the interest of remaining resources according to the model entering condition, wherein the model entering condition is a client who has full network combat and no act of remaining resources, arriving at a store, testing driving, placing orders and success;

determining the probability of having the behavioral intention of arriving at a store, testing driving, placing orders and making transactions according to the user characteristics of the customer with the reserved intention;

and determining whether the client has the intention of purchasing the vehicle conversion according to the probability.

Further, the method for determining the probability of having the behavioral intention of arriving at a store, testing driving, placing an order and being in contact is as follows:

extracting a positive sample of a rear link and a negative sample of the rear link, and obtaining user characteristics of the positive sample of the rear link and user characteristics of the negative sample of the rear link, wherein the positive sample data of the rear link is a customer with a stay behavior after full-network combat, and has a shop, test driving, ordering or success, and the negative sample data of the rear link is a customer with a stay behavior after full-network combat, but has no shop, test driving, ordering or success;

establishing a plurality of basic classifiers, dividing the combination of the user characteristics of the positive samples of the rear link and the user characteristics of the negative samples of the rear link into a rear link verification set and a rear link training set, and training each basic classifier;

fusing all the basic classifiers after training to form a final classifier;

based on the customer user characteristics with the interest in remaining resources and the final classifier, the probability that the customer has the interest in behavior to store, test drive, order placement and deal is output.

Further, the method for determining whether the customer has the intention of purchasing a vehicle is as follows:

inputting the probabilities of the behaviors of arriving at the store, trial driving, placing orders and forming the orders into an optimized rewarding function, and calculating to obtain the vehicle purchase conversion rate, wherein the independent variable of the rewarding function is the probabilities of arriving at the store, trial driving, placing orders and forming the orders, and the dependent variable is the vehicle purchase conversion rate, and the method for optimizing the rewarding function comprises the following steps:

initializing a weight value of the reward function with respect to probability of arrival at a store, trial driving, ordering and bargaining;

and taking the sample label as a state, taking a result output by a final classifier as an action, optimizing weight values of arriving shops, trial driving, placing orders and forming deals based on a Q-Learning reinforcement Learning algorithm by maximizing cumulative rewards, and obtaining an optimized rewarding function based on the optimized weight values, wherein the sample label is label numerical data and is obtained based on user characteristics of the positive sample of the rear link and user characteristics of the negative sample of the rear link.

Further, the sample tag obtaining method comprises the following steps:

and extracting clues from the user characteristics of the positive samples of the rear link and the user characteristics of the negative samples of the rear link, screening the characteristics related to the purchasing behavior from the clues, classifying and digitizing the characteristics related to the purchasing behavior according to the types of the labels, and forming the sample labels.

Further, the method for confirming the clients with the interest of remaining resources comprises the following steps:

acquiring a positive front link sample and a negative front link sample, wherein the positive front link sample is expressed as a client with a resource reserving action after full network combat, and the negative front link sample is expressed as a client without the resource reserving action after full network combat;

extracting user characteristics of a positive sample of a front link and user characteristics of a negative sample of the front link, and dividing a combination of the user characteristics of the positive sample of the front link and the user characteristics of the negative sample of the front link into a front link training set and a front link verification set;

selecting a LightGBM model to fit the training set to form a verification model, and optimizing the verification model by combining a front link verification set;

and inputting the user characteristics of the modeling condition into the optimized verification model, and outputting the clients with the reserved meanings.

An intention client acquisition system based on the automobile industry defeat clue based on the intention client acquisition method based on the automobile industry defeat clue comprises the following steps:

the condition screening module is configured to determine clients with the interest of remaining resources according to the modeling conditions, wherein the modeling conditions are clients who are full-network combat and have no actions of remaining resources, arriving at a store, testing driving, placing orders and making transactions;

the probability acquisition module is configured to acquire the user characteristics of the clients with the reserved intentions and determine the probability of the clients with the intention of going to a store, testing driving, placing orders and conducting transactions;

and the intention acquisition module is configured to determine whether the client has a purchase conversion intention according to the probability.

Further, the probability acquisition module determines the probability that the customer has the intention to go to the store, test drive, place an order and deal with the transaction by:

establishing a plurality of basic classifiers, dividing the combination of the user characteristics of the positive samples of the rear link and the user characteristics of the negative samples of the rear link into a rear link verification set and a rear link training set, and training the basic classifiers;

fusing all the basic classifiers after training to form a final classifier;

based on the customer user characteristic input with the interest in remaining resources and the final classifier, the probability that the customer has the interest in behavior to store, test drive, order placement and deal is output.

Further, the sample tag obtaining method comprises the following steps:

and extracting clues from the user characteristics of the positive samples of the rear link and the user characteristics of the negative samples of the rear link, screening the characteristics related to the shopping behavior from the clues, classifying and digitizing the characteristics related to the shopping behavior according to the types of the labels, and forming the sample labels.

Further, the method for confirming the client with the interest of remaining resources by the condition screening module comprises the following steps:

selecting a LightGBM model to fit the front link training set to form a verification model, and optimizing the verification model by combining the front link verification set;

The invention has the beneficial effects that:

the invention screens out customers with the reserve intention by re-mining and utilizing the existing defeated clue resources, analyzes the probability of having the intention of going to store, trial driving, placing orders and success according to the user characteristics of the customers, and further determines whether the customers have the intention of purchasing vehicles, so that the sales strategy can be put on targeted target customers, and data support is provided for the sales strategy appointed by the sales terminal;

the invention can help enterprises to maximally utilize the prior combat defeat clue resources, improve sales performance and customer satisfaction, reduce sales cost and marketing risk, and promote business development and data-driven decision.

Drawings

FIG. 1 is a flow chart of an embodiment;

FIG. 2 is a flow chart of the present embodiment;

fig. 3 is a structural diagram of the system.

Wherein, 1-a condition screening module; 2-a probability acquisition module; 3-intent acquisition module.

Detailed Description

Further advantages and effects of the present invention will become readily apparent to those skilled in the art from the disclosure herein, by referring to the following description of the embodiments of the present invention with reference to the accompanying drawings and preferred examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be understood that the preferred embodiments are presented by way of illustration only and not by way of limitation.

It should be noted that the illustrations provided in the following embodiments merely illustrate the basic concept of the present invention by way of illustration, and only the components related to the present invention are shown in the drawings and are not drawn according to the number, shape and size of the components in actual implementation, and the form, number and proportion of the components in actual implementation may be arbitrarily changed, and the layout of the components may be more complicated.

The method comprises two stages, wherein in the first stage, whether a user has a reserved behavior after full-network defeat is divided into a first model and whether the user has the reserved intention is judged; and the second stage is to judge the clients with the interest of remaining resources according to the first stage, and divide positive and negative samples through the behaviors of the user to store, test driving, ordering and deal with to construct a second model. The model is used for outputting the probability of the user having the behaviors of arriving at a store, testing driving, ordering and success under the premise of inputting the user characteristics of the client with the funding intention by establishing a classifier, and bringing four probability values into a reward function to determine the purchasing intention of the client.

As shown in fig. 1, the method specifically comprises:

s1: screening the modeling conditions, wherein the modeling conditions are clients who are full-network combat and have no actions of reserving resources, arriving at a store, testing driving, placing orders and completing. The embodiment aims to evaluate whether a client meeting the modeling condition has a willingness to buy a vehicle. Wherein funding means leaving personal data.

S2: and determining the clients with the reserved interests according to the modeling conditions.

In S2, a verification model needs to be built, and the first-stage model is trained by defeating data, where the data is derived from internal data and external data, specifically:

(1) internal data:

advertisement touch data: data such as advertisement reaching, clicking, watching and the like;

sales network data: the number and distribution of dealers, historical to store/deal data of dealers;

and (5) fund information: basic information of customer fund, such as province area, historical fund number, fund channel, etc.;

follow-up process data: follow-up telephone recording and voice surgery quality inspection data;

private field buried data: behavioral data on such as official platforms, APP, applet, self-building landing pages, etc.;

clue postlink data: defeat, arrival at a store, test driving, ordering, success, etc.;

sales data: sales and growth rates of different brands and different vehicle types can reflect the change of consumer demands;

production and inventory data: production quantity, productivity utilization rate, inventory turnover rate and the like, and the balance condition of production and marketing can be seen;

pre-sales communication records, whether sufficient information is provided and customer questions are answered;

historical contract records, namely whether the price and the clause meet the requirements of clients;

order log, order amount and whether the product is correct;

after-sales service record, which is to judge whether the service is timely and efficient;

customer complaints, which aspect the complaints are mainly aimed at;

(2) external data:

market research, namely, important information such as the advantages of competitors, customer satisfaction, sales volume of main competitors, price, new product release and the like.

Customer research, namely, the most interesting purchasing standard and dissatisfaction of the customer, reference data such as customer churning rate of the same industry, brand preference of the customer and the expected purchasing of the vehicle are changed.

The clue post-link data in the data are used for judging the positive and negative sample labels of the model, and the rest are used for constructing the characteristics of the model.

Data sources include customer relationship management systems (CRM) internal to the enterprise, sales management systems, market research reports, external social media, industry databases, and the like. Integrating these data into one data set facilitates subsequent data preprocessing and feature extraction.

Data cleaning: and cleaning and processing the acquired data, including data cleaning, missing value processing, abnormal value processing and the like. The data cleaning mainly removes repeated data, error data and incomplete data; the missing value processing is mainly to fill missing values, and filling is carried out by using methods of average value, median, mode and the like; outlier processing is mainly processing by removing outliers or using a scaling method or the like.

Missing value processing:

order data is missing. Some order records are lost and filled with the average or median of the historical orders.

Customer information is missing. The basic information (name, age, etc.) of some customers is missing, and the filling is performed according to the average value of the historical data.

Outlier processing:

the order amount is abnormally high. Some failed orders are of a very large amount and are corrected using the average of the historical orders.

The number of orders is abnormally large. The number of customers who run away the order at a time is very large, and abnormal orders are removed.

Customer age errors. Some customers fill out ages are very outliers, corrected using the median of the historic customer ages.

In this embodiment, a positive sample of the front link and a negative sample of the front link are extracted from the data, the positive sample of the front link is indicated as a client with a stay after full-network combat, and the negative sample of the front link is indicated as a client without a stay after full-network combat.

In this embodiment, the user characteristics of the positive sample of the front link and the user characteristics of the negative sample of the front link are extracted, the user characteristics in S2 include user attribute characteristics, network behavior characteristics, advertisement click characteristics, historical behavior characteristics, and the like, and then the processing such as characteristic crossing, conversion, standardization, and the like is performed, and the characteristic importance analysis is performed. The feature intersection can improve the nonlinear modeling capability of the model; the normalization can eliminate the influence of unit and scale differences among features, and the equal view of each dimension of features is ensured; the standardization can accelerate the solving speed of gradient descent and promote the convergence speed of the model; the feature importance analysis is to select features which are more relevant to the model target, so that the complexity of the model caused by excessive features is prevented.

In this step, the combination of the user characteristics of the positive sample and the user characteristics of the negative sample of the front link is divided into a front link training set and a front link verification set, a LightGBM model is selected to fit the front link training set, cross verification is added in the fitting process, the stability of the model is improved, a verification model is formed, the trained model is used to predict the front link verification set, and the model is continuously optimized by the effect of the model on the verification set.

In this embodiment, in this step, the user features are digitized, the user features are used as independent variables, the stay behavior and the non-stay behavior are used as dependent variables, the relationship between the user features and whether the stay behavior exists is fitted through the LightGBM model, and then whether the user features have the stay intention is judged according to the user features of the clients meeting the modeling conditions.

In this step, the user characteristics of the client conforming to the entry condition are input into the verification model, and the client having the intention to stay is output.

S3: the probability of having intent to go to the store, test drive, place order and deal is determined based on the user characteristics of the customer with intent to stay.

In this embodiment, in S3, a classifier needs to be built and trained to output probabilities of arrival at the store, test driving, placing orders, and deals.

In this embodiment, a plurality of types of basic classifiers, such as a random forest classifier, a LightGBM classifier, a neural network classifier, etc., need to be established first. When the basic classifiers are trained, the training data of each basic classifier are the same, and finally, in order to improve the output precision of the classifier, all the basic classifiers are fused to form a final classifier.

Taking LightGBM as an example, the method of training the base classifier is: setting initial parameters such as learning rate, tree depth, tree leaf number, regularization parameters, iteration times and the like, inputting a training set into a model for fitting, and optimizing according to model evaluation indexes such as AUC, accuracy, recall rate and the like until the model converges.

The data sources for training the basic classifier are:

and extracting a positive sample of the rear link and a negative sample of the rear link, and obtaining the user characteristics of the positive sample of the rear link and the user characteristics of the negative sample of the rear link, wherein the positive sample data of the rear link is a customer with a stay action after full network combat and with a shop, test driving, ordering or success, the negative sample data of the rear link is a customer with a stay action after full network combat and without a shop, test driving, ordering or success, and the data sources are as described in S2.

And then processing the positive post-link samples and the negative post-link samples, extracting the user characteristics of the positive post-link samples and the positive post-link samples, and forming sample labels, specifically: useful features are extracted from the integrated data (the back-link positive samples and the back-link negative samples), including cues that include cue sources, customer information, interaction behavior, degree of intent, and the like. And then, carrying out preliminary screening on the extracted features, selecting features related to business and application scenes, finally determining the model entering features through methods such as feature selection, correlation analysis and the like, and finally carrying out feature extraction and sample label representation by adopting deep learning models such as convolutional neural networks and the like, wherein the sample label is numerical data including the age of a client, the browsing times of certain websites, consumption records and the like.

The thread itself in this embodiment will have a thread source, and the customer information, interaction, intent will be partially internal to the host factory itself, and partially derived from external data, e.g., the trisection will have a portion of customer representation data and purchased external data.

The features related to the business and application situations are selected in the embodiment, which features are related to the purchasing behavior are judged from the business angle, and then the features are screened and classified according to the relevance of the features and the labels.

The present embodiment divides the combination of the above-described processed features into a back link verification set and a back link training set for training each base classifier.

In order to avoid the risk of overfitting of a single complex model and improve the robustness of the model, a plurality of models in the second stage are fused, and a fused prediction result is output. Common fusion methods include a weighted average method, a voting method, a fusion network, a stacking method and the like, and considering the interpretability of the fusion method and whether subjective input is needed, the stacking method is finally adopted to fuse all basic classifiers, so that a final classifier is obtained.

The specific operation process of the fusion classifier is as follows: in the second stage, a plurality of basic classifiers including random forests, lightgbms, neural networks and the like are trained firstly, and after tuning is performed through a Q-Learning algorithm, the probability values of intention of a store/test driving/ordering/success are predicted by using the tuned plurality of basic models, and the probabilities of the store, test driving, ordering and success are generated by model fusion through a stacking method.

S4: and determining whether the customer has the intention of purchasing the vehicle conversion according to the probabilities of arriving at a store, trial driving, ordering and success behavior output by the final classifier. If so, the part of clients are followed.

In this embodiment, a reward function is set, where the independent variable is the probability of the behavior of arriving at a store, driving trial, ordering and delivering, and the dependent variable is the vehicle purchase conversion rate, and then the reward function is optimized by optimizing the weight values corresponding to the store, driving trial, ordering and delivering, and finally the optimized reward function is obtained. The output probability of arriving at the store, trial driving, ordering and bargaining is input into the optimized rewarding function, the vehicle purchase conversion rate is obtained, and the vehicle purchase conversion rate is segmented according to the value of the vehicle purchase conversion rate, and in the embodiment, if the probability is more than 0.8, the vehicle purchase conversion rate is 5 minutes, namely, the vehicle purchase conversion intention is high.

For example, the optimized reward function may determine the final purchase intent using a weighted average.

The method for optimizing the reward function comprises the following steps:

in the initial state, the weight values of the behaviors of arriving at a store, testing driving, placing orders and making deals are equal, and then the weight values of the behaviors of arriving at the store, testing driving, placing orders and making deals are optimized through a Q-Learning reinforcement Learning algorithm according to the output probability values of arriving at the store, testing driving, placing orders and making deals, and the specific steps are that:

firstly, assuming equal weights of four behaviors of store/test driving/ordering/bargaining, and constructing an initial rewarding function; the result of the underlying classifier is then combined with the strategy network Q-learning using a deep reinforcement learning approach for training and optimization. The method comprises the following specific steps:

(1) Defining a state: and taking the input sample label as a state.

(2) Defining actions: taking the output probability of the classifier as an action, wherein each category corresponds to one action.

(3) Defining a reward signal: the store-arriving, test-driving, ordering and deal behaviors are respectively defined as four reward signals, each reward signal is used as a mutually independent target, and the maximization of the overall vehicle purchase conversion rate is realized by weighing a plurality of targets of store-arriving, test-driving, ordering and deal.

(4) Determining weights of the bonus signals: the weight assigned to each bonus signal is initialized.

(5) Optimizing the weights of a plurality of targets, and finding a weight distribution scheme for optimizing the vehicle purchase conversion rate:

and continuously optimizing the weight by using a Q-Learning reinforcement Learning algorithm according to the initial weight of the rewarding signal, wherein the weight of the rewarding signal is used as a state, and the increase of the vehicle purchase conversion rate is rewarded.

And deciding whether to trigger the reward signal continuously or not according to the current state and the learned weight, and updating the weight.

Through continuous iteration, the algorithm gradually finds an optimal weight distribution scheme, and then an optimized reward function is obtained.

The flow of the Q-Learning algorithm is specifically used:

the initialization, wherein an expert method gives an initial weight (state), namely the weight proportion corresponding to the current target;

selecting an Action, namely changing the weight of a certain target;

triggering Reward, namely measuring the merits of the current decision by using a Reward function, and specifically: according to the current state and the action taken, a numerical value is obtained as a reward signal of each target, and a reward is obtained by calculation according to a reward function and is used for guiding the learning and optimizing process of the reinforcement learning algorithm.

The design of the bonus function needs to be determined based on the nature and goal of the particular problem. In the scene of optimizing the vehicle purchase conversion, the rewarding function is defined according to the improvement degree of the vehicle purchase conversion, the increase of the vehicle purchase conversion is used as positive rewarding, and the decrease or no improvement of the vehicle purchase conversion is used as negative rewarding. Specific: the reward function sets an initial probability, calculates an improved vehicle purchase conversion, and compares the improved vehicle purchase conversion with the initial vehicle purchase conversion. If the improved vehicle purchase conversion rate is higher than the initial vehicle purchase conversion rate, returning the improvement degree as a forward rewards; and if the improved vehicle purchase conversion is lower than the initial vehicle purchase conversion, returning the negative number of the absolute value of the improvement degree as a negative reward.

And (3) learning, namely updating a Q table according to the report, and improving action selection.

Repeating the steps until the reward function reaches the maximum value and converges, namely finding the optimal weight proportion of the reward signal.

Training strategy network: the method comprises the steps of training a strategy network by using a near-end strategy optimization algorithm, optimizing the output probability of a classifier by maximizing accumulated rewards, and optimizing parameters of the strategy network by interaction with the environment and calculation of the accumulated rewards, so that an optimal classification strategy is realized.

Based on the above method, the flow of implementing the screening of the high-intention clients is as follows, as shown in fig. 2:

1. collecting data related to combat defeat clues;

2. data cleaning;

3. extracting features, namely extracting full-network combat data;

4. positive and negative samples of the first stage are divided into a positive sample of a front link and a negative sample of the front link;

5. training a verification model, screening clients with the reserved intent, and defining the information of the clients as medium-high intent clues;

6. dividing the middle-high intention clues into a rear link positive sample and a rear link negative sample;

7. extracting user characteristics of a back link positive sample and a back link negative sample;

8. constructing a plurality of basic classifiers;

9. fusing classifiers;

10. and screening high intention clues according to the fused classifier.

The embodiment also provides an intention client acquisition system based on the combat defeat clues of the automobile industry, as shown in fig. 3, which comprises:

the condition screening module 1 is configured to determine clients with the interest of remaining resources according to the modeling conditions, wherein the modeling conditions are clients who are full-network combat and have no activities of remaining resources, arriving at a store, testing driving, placing orders and completing;

a probability acquisition module 2 configured to acquire user characteristics of customers with interest in reservation, determine probabilities of having interest in behavior to store, test drive, order placement and deal;

the intention acquisition module 3 is configured to determine whether the customer has a intention to purchase a vehicle according to the probability.

When the system is used, the in-mold condition is input into the condition screening module 1 to obtain a customer with the intention to stay, and is input into the probability acquisition module 2, the probability acquisition module 2 outputs the user characteristics of the customer with the intention to stay, and is input into the intention acquisition module 3, and the intention acquisition module 3 determines whether the customer has the intention to purchase the vehicle.

The remaining features are described above and are not described in detail herein.

The method of the embodiment has the following advantages:

1. improving sales performance: by re-mining and utilizing existing resources of defeated cues, the model helps the enterprise to retrieve these cues and improves sales performance. The combat clues contain information such as requirements, preferences, pain points and the like of the clients, and more personalized and accurate products and services are provided for the clients through analysis of the information, so that the satisfaction degree and the loyalty degree of the clients are improved.

2. The sales cost is reduced: the cost of retooling existing defeat cues is lower than developing new customers. By utilizing the existing combat clue resources, the investment of marketing activities is reduced, and the sales efficiency and benefits are improved.

3. Optimizing marketing strategies and sales plans: the model provides advice regarding marketing strategies and sales plans to promote business development through analysis and prediction of combat cues. And marketing resources and sales resources are put on targeted target clients, so that marketing effects and sales performance are effectively improved.

4. Customer satisfaction and loyalty are improved: customer satisfaction and loyalty is enhanced by providing personalized and accurate products and services. Customer satisfaction and loyalty are important guarantees of enterprise sustainable development and are also key to improving sales performance and market competitiveness.

5. Facilitating data driven decisions: the model provides decision support based on data, helps enterprises to more scientifically and accurately make marketing strategies and sales plans, and improves the accuracy and reliability of decisions. Meanwhile, transformation and upgrading of enterprise data driving can be promoted, and the digitizing and intelligent level of enterprises can be improved.

The above embodiments are merely preferred embodiments for fully explaining the present invention, and the scope of the present invention is not limited thereto. Equivalent substitutions and modifications will occur to those skilled in the art based on the present invention, and are intended to be within the scope of the present invention.

Claims

1. The method for acquiring the intention clients based on the combat defeat clues in the automobile industry is characterized by comprising the following steps of:

2. The method for obtaining the intention clients based on the combat cues in the automotive industry according to claim 1, characterized in that: the method for determining the probability of having the behavioral intention of arriving at a store, testing driving, placing an order and conducting is as follows:

fusing all the basic classifiers after training to form a final classifier;

3. The method for obtaining the intention clients based on the combat cues in the automotive industry according to claim 2, characterized in that: the method for determining whether the client has the intention of purchasing the vehicle is as follows:

4. The method for obtaining the intention clients based on the combat cues in the automotive industry according to claim 3, characterized in that: the sample label acquisition method comprises the following steps:

5. The method for obtaining the intention clients based on the combat cues in the automotive industry according to claim 1, characterized in that: the method for confirming the clients with the interest of remaining resources comprises the following steps:

6. An intention client acquisition system based on the automobile industry defeat clue based on the intention client acquisition method based on the automobile industry defeat clue according to any one of claims 1 to 5, characterized in that: comprising the following steps:

7. The automotive industry defeat cue-based intent client acquisition system of claim 6, wherein: the probability acquisition module determines the probability that a customer has behavioral intention of arriving at a store, testing driving, placing an order and completing is as follows:

fusing all the basic classifiers after training to form a final classifier;

8. The automotive industry defeat cue-based intent client retrieval system of claim 7, wherein: the method for determining whether the client has the intention of purchasing the vehicle is as follows:

and taking the sample label as a state, taking a result output by a final classifier as an action, optimizing weight values of arriving at a store, test driving, placing an order and achieving a transaction based on a Q-Learning reinforcement Learning algorithm by maximizing cumulative rewards, and obtaining an optimized rewarding function based on the optimized weight values.

9. The automotive industry defeat cue-based intent client retrieval system of claim 8, wherein: the sample label acquisition method comprises the following steps:

10. The automotive industry defeat cue-based intent client acquisition system of claim 6, wherein: the method for confirming the clients with the funding intention by the condition screening module comprises the following steps: