Artificial intelligence system for big data analysis

Abstract

The invention belongs to the technical field of artificial intelligence, and discloses an artificial intelligence system for big data analysis. Comprising the steps of collecting customer behavior data; preprocessing the customer behavior data to obtain preprocessed behavior data; constructing a customer loss artificial intelligence prediction model, judging whether the customer is a lost customer or not according to the preprocessed behavior data and the customer loss artificial intelligence prediction model, and setting a corresponding reason label for the lost customer; making a decision according to a judgment result, wherein the decision content comprises sending a commodity coupon and carrying out commodity recommendation; according to the method, whether the customer is the lost customer is judged, and different decision methods are adopted for different customers according to the judgment result, so that the purchase rate of the commodity is improved, and the benefit of the e-commerce platform is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence system for big data analysis. Background Art

[0002] The patent application with the publication number CN117708100A discloses a business big data analysis system based on artificial intelligence, including: a data collection module for acquiring business data corresponding to an enterprise / organization; a data processing module for cleaning, deduplicating and format conversion of the acquired business data, and providing feedback on the processed data results; a data analysis module for performing statistical analysis, correlation analysis and trend analysis based on the information given in the feedback, parsing business data according to the analysis algorithm, and generating business reports. The business big data analysis system based on artificial intelligence provided by the invention greatly improves the time for system development, shortens the time required for code development, and reduces excessive consumption of manpower.

[0003] However, in the process of analyzing customer behavior big data, the data volume is large and messy, which leads to outliers in the collected customer behavior data, making it difficult to determine whether a customer is a lost customer, and the reasons for the loss of lost customers are unknown, and personalized decisions cannot be made for customers, resulting in an increase in lost customers, a decrease in the purchase rate of goods and the efficiency of e-commerce platforms.

[0004] In view of this, the present invention proposes an artificial intelligence system for big data analysis to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: an artificial intelligence system for big data analysis, comprising:

[0006] Data collection module, used to collect customer behavior data;

[0007] A data preprocessing module, used to preprocess customer behavior data to obtain preprocessed behavior data;

[0008] The churn judgment module is used to build a customer churn artificial intelligence prediction model, determine whether a customer is a churn customer based on the pre-processed behavior data and the customer churn artificial intelligence prediction model, and set corresponding reason labels for churn customers;

[0009] The decision-making module is used to make decisions based on the judgment results, including sending product coupons and making product recommendations.

[0010] Furthermore, the customer behavior data includes product browsing data, product purchase data, and product review data;

[0011] The commodity purchase data includes the time, number of times and the amount of money spent on the purchase of the commodity, and the commodity browsing data includes the number of times the commodity is browsed.

[0012] Furthermore, the method of preprocessing the customer behavior data to obtain the preprocessed behavior data includes:

[0013] Step H1: Establish a blank coordinate system, set the horizontal coordinate of the blank coordinate system to the customer's number, set the vertical coordinate of the blank coordinate system to the customer's behavior data, and fill the customer behavior data corresponding to the customer into the blank coordinate system to obtain customer behavior data sample points, and then obtain a customer behavior data distribution map;

[0014] Step H2: Obtain the Euclidean distance between each customer behavior data sample point and the remaining customer behavior data sample points, sort the Euclidean distances in ascending order, and record the remaining customer behavior data sample points corresponding to the first k Euclidean distances as neighboring behavior data sample points of the customer behavior data sample point;

[0015] Step H3: Obtain the abnormal coefficient YC(j) of the customer behavior data sample point according to the customer behavior data sample point and its corresponding neighboring behavior data sample point;

[0016] in, j is the index of the customer behavior data sample point, YC(j) is the abnormal coefficient of the j-th customer behavior data sample point, k is the number of neighboring behavior data sample points of the customer behavior data sample point, and q is the index of the neighboring behavior data sample point of the customer behavior data sample point;

[0017] in, d(j,q) is the Euclidean distance between the jth customer behavior data sample point and its corresponding qth neighboring behavior data sample point;

[0018] in, g is the index of the neighboring behavior data sample point of the qth neighboring behavior data sample point, d9jq,g) is the Euclidean distance between the qth neighboring behavior data sample point and its corresponding gth neighboring behavior data sample point;

[0019] Step H4: Set an abnormal coefficient threshold. When the abnormal coefficient of a customer behavior data sample point is greater than the abnormal coefficient threshold, record this customer behavior data sample point as an abnormal behavior data sample point. Otherwise, record it as a normal behavior data sample point. Remove the abnormal behavior data sample point and use the interpolation method to fill in the removed abnormal behavior data sample point to obtain preprocessed behavior data.

[0020] Furthermore, the method for constructing a customer churn artificial intelligence prediction model includes:

[0021] Define a churn label y for each customer;

[0022] in, t is the current time, t 近 The time when the customer purchased the product most recently from the current time, n is a positive integer;

[0023] The formula for the customer churn AI prediction model is:

[0024] Where e is the base of the natural logarithm, z represents the weighted sum of features, z = w T ×x+b; x is the input of the customer churn artificial intelligence prediction model, w T Transpose of the input to the customer churn AI prediction model, w = [w1, w2, w3]; w1, w2 and w3 are weight coefficients;

[0025] The loss function of the customer churn AI prediction model is:

[0026] Among them, m is the number of samples, i is the index of the sample, and y is the i is the loss label of sample i, p i is the predicted value of sample i, that is, the output value p(y=1|x) of the customer churn artificial intelligence prediction model; log is the logarithmic function, a is the base of the logarithmic function, a>0, and a≠1.

[0027] Furthermore, the method for determining whether a customer is a lost customer based on the pre-processed behavior data and the customer loss artificial intelligence prediction model includes:

[0028] Step A1: Collect os_r samples to form a sample set, and divide it into a training set and a validation set in a ratio of 8:2. The training set is divided into batches and input into the customer churn artificial intelligence prediction model in turn for forward propagation;

[0029] Step A2: Obtain the output value of the customer churn AI prediction model, calculate the loss value using the loss function, calculate each parameter in the model using the back propagation algorithm, and update the parameters using the gradient descent algorithm;

[0030] Step A3: Repeat A1 and A2 until the loss function value of the customer churn artificial intelligence prediction model no longer changes, import the validation set for validation, and when the validation is successful, obtain the trained customer churn artificial intelligence prediction model. When the validation is unsuccessful, repeat A3;

[0031] Step A4: construct a customer behavior matrix based on the preprocessed behavior data: kj = [kj1, kj2, kj3];

[0032] Among them, kj1 is the time difference, the time difference is: tt 近 , kj2 is the number of purchases within six months from the current time, and kj3 is the total amount of purchases within six months from the current time;

[0033] Input the customer behavior matrix into the trained customer churn artificial intelligence prediction model to obtain the output value of the customer churn artificial intelligence prediction model, and preset the output value threshold. When the output value is greater than or equal to the output value threshold, the customer is a churned customer. When the output value is less than the output value threshold, the customer is not a churned customer.

[0034] Furthermore, the method of collecting os_r samples to form a sample set includes:

[0035] Obtain os_r preprocessed behavior data and obtain the churn labels of the customers corresponding to the os_r preprocessed behavior data, construct os_r customer behavior matrices according to the os_r preprocessed behavior data, take a customer behavior matrix and the corresponding churn label as a sample, and collect all samples to form a sample set.

[0036] Furthermore, the method for presetting the output value threshold includes:

[0037] Obtain customer behavior data of D historical churn customers, construct D historical customer behavior matrices based on the customer behavior data of D historical churn customers, input the historical customer behavior matrices into the trained customer churn artificial intelligence prediction model, and obtain D historical output values;

[0038] The minimum value among the historical output values ​​is extracted as the output value threshold.

[0039] Furthermore, the method of setting corresponding reason labels for lost customers includes:

[0040] Step B1: remove punctuation marks, special symbols, numbers and blank characters in the product review data to obtain preliminary review data;

[0041] The preliminary review data is segmented using a two-way matching algorithm, all the segmented words are collected to obtain a preliminary two-way review word set, and the number of occurrences of each word in the preliminary two-way review word set is obtained;

[0042] Step B2: extract the product review data of lost customers, repeat step B1 to segment the product review data of lost customers, obtain a preliminary lost review word set, delete repeated words in the preliminary lost review word set, and obtain a final lost review word set;

[0043] Step B3: Setting a word database, wherein m_p words are stored in the word database, wherein all the words are words in the dictionary, and setting a sentiment word database based on the word database, wherein the sentiment word database is provided with a positive word storage space, a medium word storage space, and a negative word storage space, wherein positively evaluated words, medium evaluated words, and negatively evaluated words are stored respectively;

[0044] Step B4: Compare the words in the final lost comment word set with the words stored in the sentiment word database. If the word is the same as any positively evaluated word, the word is recorded as a positive word. If the word is the same as any medium-evaluated word, the word is recorded as a medium word. If the word is the same as any negatively evaluated word, the word is recorded as a negative word.

[0045] Step B5: obtaining the sentiment value of lost customers according to the preliminary two-way review word set, the preliminary lost review word set, the final lost review word set and the sentiment word database;

[0046] Step B6: Set a sentiment value threshold. When the sentiment value of the lost customer is greater than or equal to the sentiment value threshold, set the reason label of the lost customer to a positive label; otherwise, set it to a negative label.

[0047] Further, the sentiment value of the lost customer is: QG;

[0048] in,

[0049] Where α is the adjustment coefficient, α>1; s_p and SP are the index and number of positive words in the final loss comment word set, s_o and SO are the index and number of medium words in the final loss comment word set, s_r and SR are the index and number of negative words in the final loss comment word set, JF s_p 、WF s_o and QF s_r are the number of occurrences of the s_pth positive word, the s_oth medium word, and the s_rth negative word in the initial loss review word set, JZ s_p , WZ s_o and QZ s_r are the number of occurrences of the s_pth positive word, s_oth medium word and s_rth negative word in the preliminary two-way review word set, β is the adjustment coefficient, and LK is the number of product views.

[0050] Furthermore, the method for making a decision based on the judgment result includes:

[0051] When a customer is a lost customer, obtain the label of the reason for the lost customer. When the label of the reason for the lost customer is a positive label, continue to push the product to the lost customer, obtain the e-commerce platform data, which includes the product prices of competitors, the product prices on the e-commerce platform, and the product inventory quantity, and obtain the product coupons of the corresponding amount based on the e-commerce platform data, and send the product coupons to the customer;

[0052] Methods for obtaining product coupons of corresponding amounts based on e-commerce platform data include:

[0053] The formula for obtaining the product coupon amount based on the e-commerce platform data is as follows:

[0054] Among them, YH is the amount of the product coupon, M0 is the product price in the e-commerce platform, M1 is the lowest price among the competitor's product prices, zk3>zk4>zk1>zk2 and zk1, zk2, zk3 and zk4 are all greater than 0 and less than 1, K is the product inventory quantity, and L is a positive integer;

[0055] When the reason for losing a customer is labeled as a bad review, other products on the e-commerce platform are pushed to the customer;

[0056] When the customer is not a churned customer, continue to push this product to this churned customer, and simultaneously push other products on the e-commerce platform.

[0057] Technical effects and advantages of an artificial intelligence system for big data analysis of the present invention:

[0058] The customer behavior data was preprocessed to obtain the preprocessed behavior data, and it was determined whether there were outliers in the customer behavior data and the existing outliers were processed, thereby ensuring the accuracy of the subsequent judgment of whether the customer was a lost customer. Based on the preprocessed behavior data and the customer churn artificial intelligence prediction model, it was determined whether the customer was a lost customer, and corresponding reason labels were set for the lost customers. The customers were classified to make subsequent decisions more accurate, and product coupons were sent and product recommendations were made based on the judgment results, and personalized decisions were made for customers, thereby improving the retention rate of lost customers, and then increasing the purchase rate of products and the efficiency of the e-commerce platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a schematic diagram of an artificial intelligence system for big data analysis according to the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Example 1

[0062] See also Figure 1 As shown, the artificial intelligence system for big data analysis described in this embodiment includes:

[0063] Data collection module, used to collect customer behavior data;

[0064] A data preprocessing module, used to preprocess customer behavior data to obtain preprocessed behavior data;

[0065] The churn judgment module is used to build a customer churn artificial intelligence prediction model, determine whether a customer is a churn customer based on the pre-processed behavior data and the customer churn artificial intelligence prediction model, and set corresponding reason labels for churn customers;

[0066] The decision-making module is used to make decisions based on the judgment results, including sending product coupons and making product recommendations.

[0067] Methods for collecting customer behavior data include:

[0068] The customer behavior data includes product browsing data, product purchase data, and product review data;

[0069] The commodity purchase data includes the time, number of times and the amount of money spent on the purchase of the commodity, and the commodity browsing data includes the number of times the commodity is browsed;

[0070] The data collection module is composed of a customer behavior data collection terminal, and the customer behavior data collection terminal is used to collect customer behavior data;

[0071] The customer behavior data collection terminal is configured to generate a corresponding configuration channel. The server of the e-commerce platform is connected through the generated configuration channel. The customer behavior data collection terminal obtains the browsed product data, purchased product data and product review data in the server of the e-commerce platform through the corresponding configuration channel.

[0072] The customer behavior data is preprocessed, and the method for obtaining the preprocessed behavior data includes:

[0073] Step H1: Establish a blank coordinate system, set the horizontal axis of the blank coordinate system to the customer number, the customer number is "1, 2, 3, ..., h", h is the number of customers, each number corresponds to a customer, set the vertical axis of the blank coordinate system to the customer behavior data, and fill the customer behavior data corresponding to the customer into the blank coordinate system to obtain the customer behavior data sample points, and then obtain the customer behavior data distribution map;

[0074] Step H2: Obtain the Euclidean distance between each customer behavior data sample point and the remaining customer behavior data sample points, sort the Euclidean distances in ascending order, and record the remaining customer behavior data sample points corresponding to the first k Euclidean distances as neighboring behavior data sample points of the customer behavior data sample point;

[0075] The process of obtaining the Euclidean distance between each customer behavior data sample point and the remaining customer behavior data sample points includes:

[0076] Subtract the abscissa of each customer behavior data sample point from the abscissa of the remaining customer behavior data sample points, square the subtraction result, subtract the ordinate of each customer behavior data sample point from the ordinate of the remaining customer behavior data sample points, square the subtraction result, add the two square results, and perform a square root operation on the addition result to obtain the Euclidean distance between each customer behavior data sample point and the remaining customer behavior data sample points;

[0077] Step H3: obtaining the abnormal coefficient of the customer behavior data sample point according to the customer behavior data sample point and its corresponding adjacent behavior data sample point;

[0078] in, j is the index of the customer behavior data sample point, YC9j) is the abnormal coefficient of the j-th customer behavior data sample point, k is the number of neighboring behavior data sample points of the customer behavior data sample point, and q is the index of the neighboring behavior data sample point of the customer behavior data sample point;

[0079] in, d(j,q) is the Euclidean distance between the j-th customer behavior data sample point and the q-th neighboring behavior data sample point of the j-th customer behavior data sample point;

[0080] in, g is the index of the neighboring behavior data sample point of the qth neighboring behavior data sample point of the jth customer behavior data sample point, d(jq, g) is the Euclidean distance between the qth neighboring behavior data sample point of the jth customer behavior data sample point and the gth neighboring behavior data sample point of the qth neighboring behavior data sample point of the jth customer behavior data sample point;

[0081] Step H4: setting an abnormal coefficient threshold. When the abnormal coefficient of a customer behavior data sample point is greater than the abnormal coefficient threshold, the customer behavior data sample point is recorded as an abnormal behavior data sample point. Otherwise, it is recorded as a normal behavior data sample point. The abnormal behavior data sample point is removed, and the removed abnormal behavior data sample point is filled with interpolation to obtain preprocessed behavior data.

[0082] It should be explained that the above method is a process of preprocessing the browsed product data and purchased product data in the customer behavior data. The process of preprocessing the product review data is shown below.

[0083] Methods for building an AI prediction model for customer churn include:

[0084] Define a churn label y for each customer;

[0085] in, t is the current time, t 近 The time when the customer purchased the product most recently from the current time, n is a positive integer;

[0086] The formula for the customer churn AI prediction model is:

[0087] Where e is the base of the natural logarithm, z represents the weighted sum of features, z = w T ×x+b; x is the input of the customer churn artificial intelligence prediction model, w T Transpose of the input to the customer churn AI prediction model, w = [w1, w2, w3]; w1, w2 and w3 are adjustment coefficients;

[0088] The loss function of the customer churn AI prediction model is:

[0089] Among them, m is the number of samples, i is the index of the sample, and y is the i is the loss label of sample i, p i is the predicted value of sample i, that is, the output value p(y=1|x) of the customer churn artificial intelligence prediction model; log is the logarithmic function, a is the base of the logarithmic function, a>0, and a≠1;

[0090] Methods for determining whether a customer is a lost customer based on pre-processed behavioral data and the customer churn artificial intelligence prediction model include:

[0091] Step A1: Collect os_r samples to form a sample set, and divide it into a training set and a validation set in a ratio of 8:2. The training set is divided into batches and input into the customer churn artificial intelligence prediction model in turn for forward propagation;

[0092] Step A2: Obtain the output value of the customer churn AI prediction model, calculate the loss value using the loss function, calculate each parameter in the model using the back propagation algorithm, and update the parameters using the gradient descent algorithm;

[0093] Step A3: Repeat A1 and A2 until the loss function value of the customer churn artificial intelligence prediction model no longer changes, import the validation set for validation, and when the validation is successful, obtain the trained customer churn artificial intelligence prediction model. When the validation is unsuccessful, repeat A3;

[0094] Step A4: construct a customer behavior matrix based on the preprocessed behavior data: kj = [kj1, kj2, kj3];

[0095] Among them, kj1 is the time difference, the time difference is: tt 近 , kj2 is the number of purchases within six months from the current time, and kj3 is the total amount of purchases within six months from the current time;

[0096] Input the customer behavior matrix into the trained customer churn artificial intelligence prediction model to obtain the output value of the customer churn artificial intelligence prediction model, and preset the output value threshold. When the output value of the customer churn artificial intelligence prediction model is greater than or equal to the output value threshold, the customer is a churned customer; when the output value of the customer churn artificial intelligence prediction model is less than the output value threshold, the customer is not a churned customer;

[0097] The method of collecting os_r samples to form a sample set includes:

[0098] Obtain os_r preprocessed behavior data and obtain the churn labels of the customers corresponding to the os_r preprocessed behavior data, construct os_r customer behavior matrices based on the os_r preprocessed behavior data, take a customer behavior matrix and the corresponding churn label as a sample, and collect all samples to form a sample set;

[0099] Methods for presetting output value thresholds include:

[0100] Obtain customer behavior data of D historical churn customers, construct D historical customer behavior matrices based on the customer behavior data of D historical churn customers, input the historical customer behavior matrices into the trained customer churn artificial intelligence prediction model, and obtain D historical output values;

[0101] Extract the minimum value in the historical output values ​​as the output value threshold;

[0102] The process of setting corresponding reason labels for lost customers includes:

[0103] Step B1: preprocessing the product review data to obtain preliminary review data, wherein the preprocessing method includes removing punctuation marks, special symbols, numbers and blank characters in the product review data;

[0104] The preliminary review data is segmented using a two-way matching algorithm, all segmented words are collected to obtain a preliminary two-way review word set, and the number of occurrences of each word in the preliminary two-way review word set is obtained;

[0105] Step B2: extract the product review data of lost customers, repeat step B1 to pre-process and segment the product review data of lost customers, obtain a preliminary lost review word set, delete repeated words in the preliminary lost review word set, and obtain a final lost review word set;

[0106] Step B3: Setting a word database, wherein m_p words are stored in the word database, wherein all the words are words in the dictionary, and setting an emotional word database based on the word database, wherein a positive word storage space, a medium word storage space and a negative word storage space are set in the emotional word database, wherein a number of positive words are stored in the positive word storage space, such as words with positive evaluations such as "good", "great", and "perfect", a number of medium words are stored in the medium word storage space, such as words with medium evaluations such as "general", "mediocre", and a number of negative words are stored in the negative word storage space, such as words with negative evaluations such as "bad", "bad review", and "angry", and the like;

[0107] Step B4: Compare the words in the final lost comment word set with the words stored in the sentiment word database. If the word is the same as any positive word stored in the positive word storage space, the word is recorded as a positive word. If the word is the same as any medium word stored in the medium word storage space, the word is recorded as a medium word. If the word is the same as any negative word stored in the negative word storage space, the word is recorded as a negative word.

[0108] Step B5: Obtain the sentiment value QG of the lost customers according to the preliminary two-way review word set, the preliminary lost review word set, the final lost review word set and the sentiment word database;

[0109] in,

[0110] Where α is the adjustment coefficient, α>1; s_p is the index of the positive words in the final loss comment word set, SP is the number of positive words in the final loss comment word set, s_o is the index of the medium words in the final loss comment word set, SO is the number of medium words in the final loss comment word set, s_r is the index of the negative words in the final loss comment word set, SR is the number of negative words in the final loss comment word set, JF s_p is the number of occurrences of the s_pth positive word in the initial loss review word set, JZ s_p is the number of occurrences of the s_pth positive word in the initial two-way review word set, WF s_o is the number of occurrences of the s_oth medium word in the initial loss review word set, WZ s_o is the number of occurrences of the s_oth medium word in the preliminary two-way review word set, QF s_r is the number of occurrences of the s_rth negative word in the initial loss review word set, QZ s_r is the number of occurrences of the s_rth negative word in the initial two-way review word set, β is the adjustment coefficient, and LK is the number of product browsing times;

[0111] Step B6: Setting a sentiment value threshold. When the sentiment value of a lost customer is greater than or equal to the sentiment value threshold, the reason label of the lost customer is set to a positive label. When the sentiment value of a lost customer is less than the sentiment value threshold, the reason label of the lost customer is set to a negative label.

[0112] What needs to be explained is that the more times a product is browsed, the more the lost customers like the product, the higher the sentiment value of the lost customers, the more positive words and medium words there are, and the more times they appear in the preliminary two-way review word set, the higher the lost customers' evaluation of the product, the higher the sentiment value of the lost customers, the fewer negative words there are, and the fewer times they appear in the preliminary two-way review word set, the higher the sentiment value of the lost customers.

[0113] The process of making decisions based on the judgment results includes:

[0114] When a customer is a lost customer, obtain the label of the reason for the lost customer. When the label of the reason for the lost customer is a positive label, continue to push the product to the lost customer, obtain the e-commerce platform data, which includes the product prices of competitors, the product prices on the e-commerce platform, and the product inventory quantity, and obtain the product coupons of the corresponding amount based on the e-commerce platform data, and send the product coupons to the customer;

[0115] Methods for obtaining product coupons of corresponding amounts based on e-commerce platform data include:

[0116] The formula for obtaining the product coupon amount based on the e-commerce platform data is as follows:

[0117] Among them, YH is the amount of the product coupon, M0 is the product price in the e-commerce platform, M1 is the lowest price among the competitor's product prices, zk3>zk4>zk1>zk2 and zk1, zk2, zk3 and zk4 are all greater than 0 and less than 1, K is the product inventory quantity, and L is a positive integer;

[0118] When the reason for losing a customer is labeled as a bad review, other products on the e-commerce platform are pushed to the customer;

[0119] When the customer is not a churned customer, continue to push this product to this churned customer, and simultaneously push other products on the e-commerce platform.

[0120] In this embodiment, the customer behavior data is preprocessed to obtain the preprocessed behavior data, it is determined whether there are outliers in the customer behavior data, and the existing outliers are processed, thereby ensuring the accuracy of the subsequent determination of whether the customer is a lost customer, and judging whether the customer is a lost customer based on the preprocessed behavior data and the customer churn artificial intelligence prediction model, and setting corresponding reason labels for the lost customers, and classifying the customers to make subsequent decisions more accurate, and sending product coupons and making product recommendations based on the judgment results, making personalized decisions for customers, thereby improving the retention rate of lost customers, and further improving the purchase rate of products and the efficiency of the e-commerce platform.

[0121] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0122] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0123] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0124] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An artificial intelligence system for big data analysis, characterized in that: The artificial intelligence system for big data analysis includes: Data collection module, used to collect customer behavior data; A data preprocessing module, used to preprocess customer behavior data to obtain preprocessed behavior data; The churn judgment module is used to build a customer churn artificial intelligence prediction model, determine whether a customer is a churn customer based on the pre-processed behavior data and the customer churn artificial intelligence prediction model, and set corresponding reason labels for churn customers; The decision-making module is used to make decisions based on the judgment results, including sending product coupons and making product recommendations.

2. The artificial intelligence system for big data analysis according to claim 1, characterized in that: The customer behavior data includes product browsing data, product purchase data, and product review data; The commodity purchase data includes the time, number of times and the amount of money spent on the purchase of the commodity, and the commodity browsing data includes the number of times the commodity is browsed.

3. The artificial intelligence system for big data analysis according to claim 2, characterized in that: The method of preprocessing the customer behavior data to obtain the preprocessed behavior data includes: Step H1: Establish a blank coordinate system, set the horizontal coordinate of the blank coordinate system to the customer's number, set the vertical coordinate of the blank coordinate system to the customer's behavior data, and fill the customer behavior data corresponding to the customer into the blank coordinate system to obtain customer behavior data sample points, and then obtain a customer behavior data distribution map; Step H2: Obtain the Euclidean distance between each customer behavior data sample point and the remaining customer behavior data sample points, sort the Euclidean distances in ascending order, and record the remaining customer behavior data sample points corresponding to the first k Euclidean distances as neighboring behavior data sample points of the customer behavior data sample point; Step H3: Obtain the abnormal coefficient YC(j) of the customer behavior data sample point according to the customer behavior data sample point and its corresponding neighboring behavior data sample point; in, j is the index of the customer behavior data sample point, YC(j) is the abnormal coefficient of the j-th customer behavior data sample point, k is the number of neighboring behavior data sample points of the customer behavior data sample point, and q is the index of the neighboring behavior data sample point of the customer behavior data sample point; in, d(j,q) is the Euclidean distance between the jth customer behavior data sample point and its corresponding qth neighboring behavior data sample point; in, g is the index of the neighboring behavior data sample point of the qth neighboring behavior data sample point, d(jq, g) is the Euclidean distance between the qth neighboring behavior data sample point and its corresponding gth neighboring behavior data sample point; Step H4: Set an abnormal coefficient threshold. When the abnormal coefficient of a customer behavior data sample point is greater than the abnormal coefficient threshold, record this customer behavior data sample point as an abnormal behavior data sample point. Otherwise, record it as a normal behavior data sample point. Remove the abnormal behavior data sample point and use the interpolation method to fill in the removed abnormal behavior data sample point to obtain preprocessed behavior data.

4. The artificial intelligence system for big data analysis according to claim 3, characterized in that: The method for building a customer churn artificial intelligence prediction model includes: Define a churn label y for each customer; in, t is the current time, t 近 The time when the customer purchased the product most recently from the current time, n is a positive integer; The formula for the customer churn AI prediction model is: Where e is the base of the natural logarithm, z represents the weighted sum of features, z = w T ×x+b; x is the input of the customer churn artificial intelligence prediction model, w T Transpose of the input to the customer churn AI prediction model, w = [w1, w2, w3]; w1, w2 and w3 are weight coefficients; The loss function of the customer churn AI prediction model is: Among them, m is the number of samples, i is the index of the sample, and y is the i is the loss label of sample i, p i is the predicted value of sample i, that is, the output value p(y=1|x) of the customer churn artificial intelligence prediction model; log is the logarithmic function, a is the base of the logarithmic function, a>0, and a≠1.

5. The artificial intelligence system for big data analysis according to claim 4, characterized in that: The method for determining whether a customer is a lost customer based on pre-processed behavior data and a customer loss artificial intelligence prediction model includes: Step A1: Collect os_r samples to form a sample set, and divide it into a training set and a validation set in a ratio of 8:

2. The training set is divided into batches and input into the customer churn artificial intelligence prediction model in turn for forward propagation; Step A2: Obtain the output value of the customer churn AI prediction model, calculate the loss value using the loss function, calculate each parameter in the model using the back propagation algorithm, and update the parameters using the gradient descent algorithm; Step A3: Repeat A1 and A2 until the loss function value of the customer churn artificial intelligence prediction model no longer changes, import the validation set for validation, and when the validation is successful, obtain the trained customer churn artificial intelligence prediction model. When the validation is unsuccessful, repeat A3; Step A4: construct a customer behavior matrix based on the preprocessed behavior data: kj = [kj1, kj2, kj3]; Among them, kj1 is the time difference, the time difference is: tt 近 , kj2 is the number of purchases within six months from the current time, and kj3 is the total amount of purchases within six months from the current time; Input the customer behavior matrix into the trained customer churn artificial intelligence prediction model to obtain the output value of the customer churn artificial intelligence prediction model, and preset the output value threshold. When the output value is greater than or equal to the output value threshold, the customer is a churned customer. When the output value is less than the output value threshold, the customer is not a churned customer.

6. The artificial intelligence system for big data analysis according to claim 5, characterized in that: The method of collecting os_r samples to form a sample set includes: Obtain os_r preprocessed behavior data and obtain the churn labels of the customers corresponding to the os_r preprocessed behavior data, construct os_r customer behavior matrices according to the os_r preprocessed behavior data, take a customer behavior matrix and the corresponding churn label as a sample, and collect all samples to form a sample set.

7. The artificial intelligence system for big data analysis according to claim 6, characterized in that: The method for presetting the output value threshold comprises: Obtain customer behavior data of D historical churn customers, construct D historical customer behavior matrices based on the customer behavior data of D historical churn customers, input the historical customer behavior matrices into the trained customer churn artificial intelligence prediction model, and obtain D historical output values; The minimum value among the historical output values ​​is extracted as the output value threshold.

8. The artificial intelligence system for big data analysis according to claim 7, characterized in that: The method of setting a corresponding reason label for lost customers includes: Step B1: remove punctuation marks, special symbols, numbers and blank characters in the product review data to obtain preliminary review data; The preliminary review data is segmented using a two-way matching algorithm, all the segmented words are collected to obtain a preliminary two-way review word set, and the number of occurrences of each word in the preliminary two-way review word set is obtained; Step B2: extract the product review data of lost customers, repeat step B1 to segment the product review data of lost customers, obtain a preliminary lost review word set, delete repeated words in the preliminary lost review word set, and obtain a final lost review word set; Step B3: Setting a word database, wherein m_p words are stored in the word database, wherein all the words are words in the dictionary, and setting a sentiment word database based on the word database, wherein the sentiment word database is provided with a positive word storage space, a medium word storage space, and a negative word storage space, wherein positively evaluated words, medium evaluated words, and negatively evaluated words are stored respectively; Step B4: Compare the words in the final lost comment word set with the words stored in the sentiment word database. If the word is the same as any positively evaluated word, the word is recorded as a positive word. If the word is the same as any medium-evaluated word, the word is recorded as a medium word. If the word is the same as any negatively evaluated word, the word is recorded as a negative word. Step B5: obtaining the sentiment value of lost customers according to the preliminary two-way review word set, the preliminary lost review word set, the final lost review word set and the sentiment word database; Step B6: Set a sentiment value threshold. When the sentiment value of the lost customer is greater than or equal to the sentiment value threshold, set the reason label of the lost customer to a positive label; otherwise, set it to a negative label.

9. The artificial intelligence system for big data analysis according to claim 8, characterized in that: The sentiment value of the lost customer is: QG; in, Where α is the adjustment coefficient, α>1; s_p and SP are the index and number of positive words in the final loss comment word set, s_o and SO are the index and number of medium words in the final loss comment word set, s_r and SR are the index and number of negative words in the final loss comment word set, JF s_p 、WF s_o and QF s_r are the number of occurrences of the s_pth positive word, the s_oth medium word, and the s_rth negative word in the initial loss review word set, JZ s_p , WZ s_o and QZ s_r are the number of occurrences of the s_pth positive word, s_oth medium word and s_rth negative word in the preliminary two-way review word set, β is the adjustment coefficient, and LK is the number of product views.

10. An artificial intelligence system for big data analysis according to claim 9, characterized in that: The method for making a decision based on the judgment result comprises: When a customer is a lost customer, obtain the label of the reason for the lost customer. When the label of the reason for the lost customer is a positive label, continue to push the product to the lost customer, obtain the e-commerce platform data, which includes the product prices of competitors, the product prices on the e-commerce platform, and the product inventory quantity, and obtain the product coupons of the corresponding amount based on the e-commerce platform data, and send the product coupons to the customer; Methods for obtaining product coupons of corresponding amounts based on e-commerce platform data include: The formula for obtaining the product coupon amount based on the e-commerce platform data is: YH = Among them, YH is the amount of the product coupon, M0 is the product price in the e-commerce platform, M1 is the lowest price among the competitor's product prices, zk3>zk4>zk1>zk2 and zk1, zk2, zk3 and zk4 are all greater than 0 and less than 1, K is the product inventory quantity, and L is a positive integer; When the reason for losing a customer is labeled as a bad review, other products on the e-commerce platform are pushed to the customer; When the customer is not a churned customer, continue to push this product to this churned customer, and simultaneously push other products on the e-commerce platform.

Images