Customer demand prediction system and method based on machine learning

Abstract

The invention belongs to the technical field of customer demand prediction, and particularly relates to a customer demand prediction system and method based on machine learning. According to the method, the accuracy of demand prediction is effectively improved by introducing a multi-dimensional fusion prediction method of historical demands, real-time behavior characteristics and preference information, abnormal changes of customer behaviors can be quickly responded through deviation behavior recognition, customer demands in a dynamic environment are adapted, and the accuracy of demand prediction is improved by dividing the first time period and the second time period. According to the method, real-time dynamic changes of clients are captured, prediction of long-term demand preferences is considered, scientificity and comprehensiveness of results are improved, enterprises can provide more targeted and personalized services through accurate prediction of client demands, and client experience and satisfaction are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of customer demand prediction, and specifically relates to a customer demand prediction system and method based on machine learning. Background Art

[0002] As market competition intensifies, companies are paying more and more attention to improving the personalization of products and services through accurate demand forecasting, so as to better meet customer needs, enhance customer experience and optimize resource allocation. Traditional customer demand forecasting methods rely on historical data analysis and simple statistical models. These methods can only predict customer needs based on fixed rules or trends, ignoring the dynamic changes and individual differences of customer needs. Therefore, traditional methods are difficult to cope with complex and changing market environments, especially in large customer groups and rapidly changing market conditions, and there are large forecast errors and response lags.

[0003] In recent years, the application of machine learning technology in various industries has continued to deepen, especially in customer behavior analysis and demand forecasting. The customer demand forecasting method based on machine learning can explore the potential laws in customer behavior and accurately predict customer demand through in-depth analysis of large amounts of data.

[0004] However, each customer has large individual differences in demand preferences, behavioral habits, etc. Traditional demand forecasting methods are often unable to make accurate personalized forecasts for different customer groups, resulting in forecast results that are too rough and unable to effectively meet customers' actual needs. Summary of the invention

[0005] The purpose of this invention is to provide a customer demand forecasting method based on machine learning, which can improve forecasting accuracy, adapt to the rapid changes in market demand, and provide enterprises with accurate personalized service recommendations.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A customer demand prediction method based on machine learning, comprising:

[0008] Obtain customer behavior data, and obtain current behavior characteristics information based on the behavior data;

[0009] Obtain the customer's historical demand information, obtain demand similarity information based on the current behavior characteristic information and the customer's historical demand information, and determine whether the demand similarity information meets the first preset condition. If not, mark it as deviant behavior information;

[0010] Construct the first period and obtain customer demand trend information within the first period;

[0011] Constructing a second time period, obtaining customer preference data in the second time period, and obtaining corresponding preference information according to the customer preference data;

[0012] Obtain customer demand behavior information based on deviation behavior information, demand trend information and preference information, and obtain customer demand information based on customer demand behavior information.

[0013] In a preferred solution, the steps of obtaining historical demand information of a customer, obtaining demand similarity information according to current behavior characteristic information and historical demand information of the customer, and judging whether the demand similarity information meets the first preset condition, and if not, marking it as deviation behavior information, include:

[0014] Obtain customer historical demand information;

[0015] Obtain corresponding multiple historical behavior characteristic information based on the customer's historical demand information;

[0016] Obtain the historical behavior characteristic vector corresponding to each historical behavior characteristic information;

[0017] Obtaining a corresponding current behavior feature vector according to the current behavior feature information;

[0018] Obtain demand similarity values ​​according to historical behavior feature vectors and current behavior feature vectors, and mark them as demand similarity information;

[0019] It is determined whether the demand similarity information meets the first preset condition. If not, it is marked as deviant behavior information.

[0020] In a preferred solution, the step of determining whether the demand similarity information meets the first preset condition, and if not, marking it as deviation behavior information, includes:

[0021] Get the demand similarity threshold;

[0022] Determine whether the demand similarity value exceeds the demand similarity threshold;

[0023] If all demand similarity values ​​do not exceed the demand similarity threshold, it is determined that the current behavior characteristic information does not match the customer's historical demand information, and the current behavior characteristic information is marked as deviant behavior information.

[0024] In a preferred solution, the step of constructing a first time period and obtaining customer demand trend information within the first time period includes:

[0025] Construct the first period;

[0026] Obtaining first customer demand data in a first period;

[0027] Acquire a plurality of corresponding first customer demand vectors according to the first customer demand data;

[0028] A demand trend value is obtained according to the first vectors of multiple customer demands, and is marked as customer demand trend information.

[0029] In a preferred embodiment, the step of constructing the first time period includes:

[0030] Obtaining the time point marked as deviant behavior information and marking it as the end time of the first period;

[0031] Obtain multiple demand similarity values ​​and summarize them to obtain a demand similarity table;

[0032] Obtain the upper limit value in the demand similarity table and mark it as the maximum demand similarity value;

[0033] Obtain a duration table, wherein the duration table includes a plurality of demand similarity interval values ​​and a first duration corresponding to each demand similarity interval value;

[0034] Obtain the target demand similarity interval value according to the maximum demand similarity value;

[0035] Obtain the corresponding first duration from the duration table according to the target demand similarity interval;

[0036] Obtaining the start time of the first period according to the first duration and the end time of the first period;

[0037] The first time period is obtained according to the start time of the first time period and the end time of the first time period.

[0038] In a preferred solution, the steps of constructing a second time period, obtaining customer preference data in the second time period, and obtaining corresponding preference information according to the customer preference data include:

[0039] Construct the second period;

[0040] Obtain customer preference data in the second period;

[0041] Acquire corresponding multiple customer preference items according to the customer preference data;

[0042] Obtain multiple customer preference vectors corresponding to each customer preference item;

[0043] A customer preference coefficient is obtained according to a plurality of customer preference items and a plurality of corresponding customer preference vectors, and is marked as preference information.

[0044] In a preferred embodiment, the step of constructing the second time period includes:

[0045] Obtaining the time point marked as deviant behavior information and marking it as the end time of the second period;

[0046] Get the first duration of the first time period;

[0047] Obtain standard customer preference coefficients;

[0048] Obtain a second duration according to a customer preference coefficient, a standard customer preference coefficient, and a first duration;

[0049] Obtaining the start time of the second period according to the second duration and the end time of the second period;

[0050] The second time period is obtained according to the start time of the second time period and the end time of the second time period.

[0051] In a preferred embodiment, the step of obtaining customer demand behavior information based on the deviation behavior information, demand trend information and preference information, and obtaining customer demand information based on the customer demand behavior information includes:

[0052] Obtaining the corresponding current behavior feature vector according to the deviation behavior information;

[0053] Obtain the corresponding standard behavior feature vector based on customer behavior data;

[0054] Obtain the customer demand value based on the current behavior characteristic vector, demand trend information, and preference information, and mark it as customer demand behavior information;

[0055] Obtaining a demand forecast table, wherein the demand forecast table includes a plurality of customer demand interval values ​​and customer demand information corresponding to each demand interval value;

[0056] Obtain the corresponding target customer demand interval value according to the customer demand value;

[0057] Obtain the corresponding customer demand information from the demand forecast table based on the target customer demand range value.

[0058] The present invention also provides a customer demand prediction system based on machine learning, which is used in the above-mentioned customer demand prediction method based on machine learning, comprising:

[0059] The behavior characteristics module is used to obtain customer behavior data and obtain current behavior characteristics information based on the behavior data;

[0060] A behavior judgment module is used to obtain historical demand information of customers, obtain demand similarity information based on current behavior characteristic information and historical demand information of customers, and judge whether the demand similarity information meets the first preset condition. If not, it is marked as deviation behavior information;

[0061] The first time period module is used to construct the first time period and obtain customer demand trend information within the first time period;

[0062] The second time period module is used to construct the second time period, obtain the customer preference data in the second time period, and obtain the corresponding preference information according to the customer preference data;

[0063] The demand matching module is used to obtain customer demand behavior information based on deviation behavior information, demand trend information and preference information, and obtain customer demand information based on customer demand behavior information.

[0064] And, a customer demand prediction terminal based on machine learning, comprising:

[0065] one or more processors;

[0066] a storage device having one or more programs stored thereon;

[0067] When one or more programs are executed by one or more processors, the one or more processors implement a customer demand prediction method based on machine learning.

[0068] The technical effects achieved by the present invention are:

[0069] The present invention effectively improves the accuracy of demand forecasting by introducing a multi-dimensional fusion prediction method of historical demand, real-time behavioral characteristics and preference information, can quickly respond to abnormal changes in customer behavior through deviation behavior identification, and adapt to customer needs in a dynamic environment. By dividing the first time period and the second time period, it not only captures the real-time dynamic changes of customers, but also takes into account the prediction of long-term demand preferences, thereby improving the scientificity and comprehensiveness of the results. By accurately predicting customer needs, enterprises can provide more targeted and personalized services and enhance customer experience and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is a flow chart of the method provided by the present invention;

[0071] Figure 2 It is a system module diagram provided by the present invention. DETAILED DESCRIPTION

[0072] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0073] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0074] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" that appears in different places in this specification does not refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0075] Secondly, the present invention is described in detail in conjunction with schematic diagrams. When describing the embodiments of the present invention in detail, for the convenience of explanation, the schematic diagrams are only examples and should not limit the scope of protection of the present invention.

[0076] Please see attached Figure 1 As shown, a customer demand prediction method based on machine learning is provided, including:

[0077] S1. Obtain customer behavior data and obtain current behavior characteristics information based on the behavior data;

[0078] S2. Obtain the customer's historical demand information, obtain demand similarity information based on the current behavior characteristic information and the customer's historical demand information, and determine whether the demand similarity information meets the first preset condition. If not, mark it as deviation behavior information;

[0079] S3, constructing a first period, and obtaining customer demand trend information within the first period;

[0080] S4. Construct a second time period, obtain customer preference data in the second time period, and obtain corresponding preference information according to the customer preference data;

[0081] S5. Obtain customer demand behavior information based on the deviation behavior information, demand trend information and preference information, and obtain customer demand information based on the customer demand behavior information.

[0082] As in the above steps S1 to S5, by collecting the customer's real-time behavior data, extracting behavior characteristic information, such as visit duration, purchase frequency, browsing category, etc., to form a current behavior characteristic vector, obtain the customer's historical demand information (such as past consumption records and behavior characteristics), and match it with the current behavior characteristics, calculate demand similarity information, and if the demand similarity information does not reach the preset similarity threshold, it is marked as deviation behavior information, indicating that the customer's current demand is greatly deviated from its historical demand. In the first time period (from a specific time point to the current short-term time window), collect customer demand data and generate customer demand trend information. In the second time period (a longer time window), analyze the customer's preference data, such as preference category, price sensitivity etc., extract preference information, integrate deviation behavior information, demand trend information and preference information, generate customer demand behavior information, and obtain the final customer demand information based on demand forecasts. By introducing a multi-dimensional fusion forecasting method of historical demand, real-time behavior characteristics and preference information, the accuracy of demand forecasts is effectively improved. It can quickly respond to abnormal changes in customer behavior through deviation behavior identification and adapt to customer needs in a dynamic environment. By dividing the first time period (short-term) and the second time period (long-term), it captures the real-time dynamic changes of customers and takes into account the prediction of long-term demand preferences, which improves the scientificity and comprehensiveness of the results. By accurately predicting customer needs, companies can provide more targeted and personalized services and enhance customer experience and satisfaction.

[0083] In a preferred embodiment, the steps of obtaining historical demand information of a customer, obtaining demand similarity information according to current behavior characteristic information and historical demand information of the customer, and determining whether the demand similarity information meets a first preset condition, and if not, marking it as deviation behavior information, include:

[0084] S201, obtaining historical demand information of customers;

[0085] S202, obtaining corresponding multiple historical behavior characteristic information according to the customer's historical demand information;

[0086] S203, obtaining a historical behavior characteristic vector corresponding to each historical behavior characteristic information;

[0087] S204, acquiring a corresponding current behavior characteristic vector according to the current behavior characteristic information;

[0088] S205, obtaining a demand similarity value according to the historical behavior characteristic vector and the current behavior characteristic vector, and marking it as demand similarity information;

[0089] S206: Determine whether the demand similarity information meets the first preset condition. If not, mark it as deviation behavior information.

[0090] As in the above steps S201 to S206, historical demand information is extracted from the customer's past behavior records, including the types of purchased goods, service usage, behavior patterns, etc. Based on the customer's historical demand information, multiple historical behavior characteristics are extracted, such as purchase frequency, average consumption amount, common visit time period, etc., and multiple historical behavior characteristic sets are generated to describe the customer's historical behavior pattern. Each historical behavior characteristic is converted into a mathematical vector form (historical behavior characteristic vector), and the customer's current behavior characteristic information (such as recent visit and interaction data) is feature extracted and quantized to generate a current behavior characteristic vector, which is uniformly standardized with historical data. Based on the historical behavior characteristic vector and the current behavior characteristic vector, a demand similarity value is calculated to reflect the degree of matching between the customer's current behavior and its historical behavior pattern, which is marked as demand similarity information. The calculation formula for the demand similarity value is: Where, X demand It is represented as the demand similarity value, A is represented as the current behavior characteristic vector, and B is represented as the historical behavior characteristic vector. The demand similarity value is compared with the preset similarity threshold (the first preset condition). If it does not meet the requirements, it is marked as deviation behavior information. It can quickly identify anomalies or deviations in customer behavior, provide real-time early warning for enterprises, help adjust strategies or recommend content, fully tap the value of customers' historical data, combine current behavior characteristics to achieve accurate matching, and provide comprehensive data support for demand forecasting.

[0091] In a preferred embodiment, the step of determining whether the demand similarity information meets the first preset condition, and if not, marking it as deviation behavior information, includes:

[0092] S2061. Obtaining a demand similarity threshold;

[0093] S2062, determining whether the demand similarity value exceeds the demand similarity threshold;

[0094] If all demand similarity values ​​do not exceed the demand similarity threshold, it is determined that the current behavior characteristic information does not match the customer's historical demand information, and the current behavior characteristic information is marked as deviant behavior information.

[0095] As in the above steps S2061 to S2062, according to the specific application scenario, a predetermined demand similarity threshold is set as a standard for measuring whether the customer's current behavior is consistent with the historical behavior pattern. The threshold can be set based on historical data statistics, expert experience or algorithm optimization. The threshold can be dynamically adjusted according to different times, scenarios or customer groups (such as VIP customers and ordinary customers) to improve adaptability. According to the multiple demand similarity values ​​calculated previously, they are compared with the demand similarity threshold one by one. If any demand similarity value exceeds the demand similarity threshold, it is considered that the current behavior characteristics are consistent with at least one historical demand pattern and are not considered as deviation behavior. If all demand similarity values ​​are lower than the demand similarity threshold, it indicates that the current behavior characteristics are completely inconsistent with the historical pattern, indicating that the customer behavior may have been abnormal or new demand preferences have emerged. The current behavior characteristics information is marked as deviation behavior information. By setting the demand similarity threshold, it can effectively distinguish between normal and abnormal customer behavior, helping enterprises to more accurately locate behavioral deviations. By marking deviation behavior information, enterprises can capture possible changes in customer interests or demand upgrades in real time, providing a basis for formulating personalized service strategies.

[0096] In a preferred embodiment, the step of constructing a first time period and obtaining customer demand trend information within the first time period includes:

[0097] S301, constructing a first period;

[0098] S302, obtaining first customer demand data within a first period;

[0099] S303, acquiring corresponding multiple first customer demand vectors according to the first customer demand data;

[0100] S304: Obtain demand trend values ​​according to multiple customer demand first vectors, and mark them as customer demand trend information.

[0101] As in the above steps S301 to S304, time is divided into specific analysis segments based on the time series data of customer behavior. The first time period can be dynamically set based on the significant change points of historical behavior data or the time segmentation requirements of business scenarios. The start and end times of the first time period are determined based on the results of behavior deviation information, demand similarity analysis, etc., to ensure that the selected time period can represent the characteristics of recent customer demand changes. All relevant behavior data of customers in the first time period, such as access records, purchase records, search keywords, etc., are collected, and invalid or abnormal data are eliminated to ensure data quality to improve the accuracy of subsequent analysis. The customer demand data collected in the first time period is converted into a standardized multidimensional vector (such as demand type, frequency, preference weight, etc.) through a feature extraction algorithm. Each customer demand first vector represents a behavior fragment, and the complete vector set is used to represent the customer's demand structure and preference characteristics in the first time period. Using multiple customer demand first vectors, the demand trend value is calculated and marked as customer demand trend information. The calculation formula of the demand trend value is: In the formula, Q trend It is represented as the demand trend value, i represents the first vector of multiple customer demands, i = 2, 3, 4…n, Q i It is represented as the first vector of the i-th customer demand, Q i-1 It is represented as the first vector of the i-1th customer demand. Through time segmentation and demand trend extraction, it can capture the dynamic changes of customer demand in a specific time period, help enterprises better understand customer behavior patterns, and ensure that the prediction results are based on the latest customer behavior data, thereby improving the accuracy and timeliness of the prediction.

[0102] In a preferred embodiment, the step of constructing the first time period includes:

[0103] S3011, obtaining a time point marked as deviant behavior information, and marking it as the end time of the first period;

[0104] S3012, obtaining multiple demand similarity values, and summarizing them to obtain a demand similarity table;

[0105] S3013, obtaining the upper limit value in the demand similarity table and marking it as the maximum demand similarity value;

[0106] S3014, obtaining a duration table, wherein the duration table includes a plurality of demand similarity interval values ​​and a first duration corresponding to each demand similarity interval value;

[0107] S3015. Obtain a target demand similarity interval value according to the maximum demand similarity value;

[0108] S3016, obtaining a corresponding first duration from the duration table according to the target demand similarity interval;

[0109] S3017, obtaining the start time of the first time period according to the first duration and the end time of the first time period;

[0110] S3018. Obtain the first time period according to the start time of the first time period and the end time of the first time period.

[0111] As in the above steps S3011 to S3018, the time point marked as the deviation behavior information is obtained, and the time point is determined to be the end time of the first time period. Multiple demand similarity values ​​calculated in the historical data are collected, and a demand similarity table is generated. The upper limit value (maximum demand similarity value) is extracted from the demand similarity table to indicate the maximum matching degree of customer demand. The predefined duration table is loaded. The duration table associates the demand similarity interval value and the analysis period length (first duration) corresponding to each interval value. The table is constructed based on actual business needs or historical analysis experience and is used for dynamic adjustment of the time period range. According to the maximum demand similarity value, the demand similarity interval value is located. , according to the matched demand similarity interval value, the corresponding first duration is obtained from the duration table, the first duration is combined with the end time of the first time period, and the start time of the first time period is calculated forward to form a complete time period range, and the first time period is finally determined according to the start time and end time, providing a time period data basis for subsequent customer demand trend analysis. By combining the demand similarity value with the duration table, the time period range can be dynamically adjusted to make the time period selection more flexible and accurate, thereby capturing the demand changes in the key time period, and determining the time period range according to the demand similarity and deviation behavior information, avoiding the interference of irrelevant or redundant data, and optimizing data processing efficiency.

[0112] In a preferred embodiment, the steps of constructing a second time period, obtaining customer preference data within the second time period, and obtaining corresponding preference information according to the customer preference data include:

[0113] S401, constructing a second period;

[0114] S402, obtaining customer preference data in a second period;

[0115] S403, acquiring corresponding multiple customer preference items according to the customer preference data;

[0116] S404, obtaining multiple customer preference vectors corresponding to each customer preference item;

[0117] S405 . Obtain customer preference coefficients according to multiple customer preference items and their corresponding multiple customer preference vectors, and mark them as preference information.

[0118] As in the above steps S401 to S405, a second time period is constructed based on business needs or time rules. This time period is determined in order to analyze changes in customer preferences in different time periods, and obtain customer preference data in the second time period, which usually includes customer purchase history, browsing history, interactive behavior, etc. These data reflect the customer's long-term and recent preferences. The customer's preference data is further subdivided into multiple customer preference items. For example, customers may have different areas of interest (such as product categories, brand preferences, etc.), and these specific preference items need to be identified and extracted for in-depth analysis. Multiple customer preference vectors are constructed for each preference item. These vectors usually represent the performance characteristics of the customer under a specific item, such as interest intensity, purchase frequency, etc. Through these vectors, the customer's tendency and interest level for different preferences can be quantified. Using multiple customer preference items and their corresponding preference vectors, the customer preference coefficient is obtained by calculation. The calculation formula of the customer preference coefficient is: In the formula, G like is represented by the customer preference coefficient, k is represented by the number of multiple customer preference items, k = 1, 2, 3 ... m, g is represented by the number of multiple customer preference vectors corresponding to each customer preference item, where g = 1, 2, 3 ... j, G g It is represented as the g-th customer preference vector corresponding to each customer preference item. The acquisition of preference coefficients can provide support for personalized recommendations.

[0119] In a preferred embodiment, the step of constructing the second time period includes:

[0120] S4011, obtaining a time point marked as deviant behavior information, and marking it as the end time of the second period;

[0121] S4012, obtaining a first duration of a first time period;

[0122] S4013. Obtain standard customer preference coefficient;

[0123] S4014. Obtain a second duration according to the customer preference coefficient, the standard customer preference coefficient, and the first duration;

[0124] S4015, obtaining the start time of the second time period according to the second duration and the end time of the second time period;

[0125] S4016. Obtain the second time period according to the start time of the second time period and the end time of the second time period.

[0126] As in the above steps S4011 to S4016, in the first time period, the marked deviation behavior information is the abnormal point of customer demand or behavior. According to these abnormal points, the end time of the second time period can be set to obtain the first duration of the first time period. The standard customer preference coefficient is a standard coefficient defined based on the common behavior or preference data of a large number of customers, which is used to measure the average preference of customers for certain products or services. The duration of the second time period is calculated by combining the actual preference coefficient of the customer, the standard preference coefficient and the duration of the first time period. The calculation formula of the second duration is: Where, T second Expressed as the second duration, T first Represented as the first duration, G like Expressed as customer preference coefficient, G standard Expressed as a standard customer preference coefficient, the length of the second time period can be dynamically adjusted to reflect the actual changes in customer demand under specific preference conditions. The start time of the second time period can be reversely calculated through the length of the second time period and the set end time. Combined with the end time of the first time period and the calculated start time and end time of the second time period, the second time period can be accurately constructed, providing a data basis for the analysis of customer needs and preferences, ensuring that the division of the second time period is more in line with the actual changes in customer demand, avoiding overly rough time period division, and improving the accuracy of demand forecasting. According to the deviation behavior of the first time period and the customer preference coefficient, the length of the second time period can be dynamically adjusted to flexibly respond to changes in customer demand, so that fluctuations in customer preferences and changes in demand can be captured in a timely manner, improving the timeliness and accuracy of forecasting.

[0127] In a preferred embodiment, the step of obtaining customer demand behavior information based on the deviation behavior information, demand trend information and preference information, and obtaining customer demand information based on the customer demand behavior information includes:

[0128] S501, obtaining a corresponding current behavior characteristic vector according to the deviation behavior information;

[0129] S502, obtaining a corresponding standard behavior characteristic vector according to the customer behavior data;

[0130] S503, obtaining a customer demand value according to the current behavior characteristic vector, demand trend information, and preference information, and marking it as customer demand behavior information;

[0131] S504, obtaining a demand forecast table, wherein the demand forecast table includes a plurality of customer demand interval values ​​and customer demand information corresponding to each demand interval value;

[0132] S505, obtaining a corresponding target customer demand interval value according to the customer demand value;

[0133] S506. Obtain corresponding customer demand information from the demand forecast table according to the target customer demand interval value.

[0134] As in the above steps S501 to S506, by obtaining the time point marked as the deviation behavior information, it is matched with the current customer behavior data to generate the current behavior characteristic vector, and by collecting the customer's historical behavior data, a standard behavior characteristic vector is generated. The standard behavior characteristic vector represents the typical behavior pattern of the customer, which is the behavioral characteristics exhibited by the customer under normal circumstances. By combining the current behavior characteristic vector, demand trend information and preference information, the customer demand value of the current period is calculated and marked as "customer demand behavior information". The calculation formula of the customer demand value is: In the formula, Z represents the customer demand value, A represents the current behavior feature vector, and Q trend Expressed as demand trend value, G like Expressed as the customer preference coefficient, C as the standard behavior characteristic vector, the demand forecast table is called, which includes multiple customer demand interval values, each demand interval value corresponds to different customer demand information. The demand forecast table is based on historical data and machine learning algorithms, and can predict customer demand information corresponding to different demand intervals. For example, the demand interval may be divided into low, medium, and high demand intervals, and each interval corresponds to a different degree of demand and potential product demand. According to the customer demand value calculated earlier, the customer demand value is matched with the demand interval in the demand forecast table to find the demand interval that best matches the customer demand. According to the target demand interval value corresponding to the customer demand value, the corresponding customer demand information is extracted from the demand forecast table. The demand information may include specific product recommendations, service suggestions or other personalized services that match customer needs. Through this process, specific and personalized demand forecast information can be provided to ensure the accurate matching of customer needs, so that the forecast results are more in line with the actual needs of customers.

[0135] Please see attached Figure 2 As shown, the present invention also provides a customer demand prediction system based on machine learning, which is used in the above-mentioned customer demand prediction method based on machine learning, comprising:

[0136] The behavior characteristics module is used to obtain customer behavior data and obtain current behavior characteristics information based on the behavior data;

[0137] A behavior judgment module is used to obtain historical demand information of customers, obtain demand similarity information based on current behavior characteristic information and historical demand information of customers, and judge whether the demand similarity information meets the first preset condition. If not, it is marked as deviation behavior information;

[0138] The first time period module is used to construct the first time period and obtain customer demand trend information within the first time period;

[0139] The second time period module is used to construct the second time period, obtain the customer preference data in the second time period, and obtain the corresponding preference information according to the customer preference data;

[0140] The demand matching module is used to obtain customer demand behavior information based on deviation behavior information, demand trend information and preference information, and obtain customer demand information based on customer demand behavior information.

[0141] As mentioned above, the behavior characteristics module obtains the customer's behavior data and extracts the current behavior characteristics information based on these data. The behavior data includes but is not limited to the customer's purchase record, browsing record, usage time, interaction method, etc. Through these behavior data, the customer's activity characteristics in a specific period of time can be obtained to form the customer's behavior characteristics information. The behavior judgment module obtains the customer's historical demand information and combines it with the current behavior characteristics information to obtain demand similarity. If the demand similarity does not meet the preset conditions, the current behavior will be marked as "deviant behavior information". The first time period module constructs a "first time period" based on the customer's historical demand information and behavior characteristics. This time period covers the customer's past demand change trend and obtains customer demand trend information. The second time period module obtains the customer's preference data in the second time period and extracts the customer's preference information based on these data. The demand matching module integrates the data provided by the previous modules, including deviation behavior information, demand trend information and preference information, to infer the customer's demand behavior. Finally, through the customer demand behavior information, the customer's demand information is obtained, which can provide more accurate and personalized demand forecasts. By identifying deviation behaviors and inferring demand trends, it can provide customer demand forecasts with higher accuracy, reduce resource waste, and improve customer satisfaction.

[0142] And, a customer demand prediction terminal based on machine learning, comprising:

[0143] one or more processors;

[0144] a storage device having one or more programs stored thereon;

[0145] When one or more programs are executed by one or more processors, the one or more processors implement a customer demand prediction method based on machine learning.

[0146] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. A customer demand prediction method based on machine learning, characterized in that: include: Obtain customer behavior data, and obtain current behavior characteristics information based on the behavior data; Obtain the customer's historical demand information, obtain demand similarity information based on the current behavior characteristic information and the customer's historical demand information, and determine whether the demand similarity information meets the first preset condition. If not, mark it as deviant behavior information; Construct the first period and obtain customer demand trend information within the first period; Constructing a second time period, obtaining customer preference data in the second time period, and obtaining corresponding preference information according to the customer preference data; Obtain customer demand behavior information based on deviation behavior information, demand trend information and preference information, and obtain customer demand information based on customer demand behavior information.

2. The method for predicting customer demand based on machine learning according to claim 1, characterized in that: The steps of obtaining historical demand information of the customer, obtaining demand similarity information according to the current behavior characteristic information and the historical demand information of the customer, and judging whether the demand similarity information meets the first preset condition, and if not, marking it as deviant behavior information, include: Obtain customer historical demand information; Obtain corresponding multiple historical behavior characteristic information based on the customer's historical demand information; Obtain the historical behavior characteristic vector corresponding to each historical behavior characteristic information; Obtaining a corresponding current behavior feature vector according to the current behavior feature information; Obtain demand similarity values ​​according to historical behavior characteristic vectors and current behavior characteristic vectors, and mark them as demand similarity information; It is determined whether the demand similarity information meets the first preset condition. If not, it is marked as deviant behavior information.

3. The method for predicting customer demand based on machine learning according to claim 2, characterized in that: The step of determining whether the demand similarity information meets the first preset condition, and if not, marking it as deviant behavior information, comprises: Get the demand similarity threshold; Determine whether the demand similarity value exceeds the demand similarity threshold; If all demand similarity values ​​do not exceed the demand similarity threshold, it is determined that the current behavior characteristic information does not match the customer's historical demand information, and the current behavior characteristic information is marked as deviant behavior information.

4. The method for predicting customer demand based on machine learning according to claim 2, characterized in that: The steps of constructing the first period and obtaining the customer demand trend information in the first period include: Construct the first period; Obtaining first customer demand data in a first period; Acquire a plurality of corresponding first customer demand vectors according to the first customer demand data; A demand trend value is obtained according to a plurality of customer demand first vectors and marked as customer demand trend information.

5. The method for predicting customer demand based on machine learning according to claim 4, characterized in that: The steps to construct the first period include: Obtaining the time point marked as deviant behavior information and marking it as the end time of the first period; Obtain multiple demand similarity values ​​and summarize them to obtain a demand similarity table; Obtain the upper limit value in the demand similarity table and mark it as the maximum demand similarity value; Obtain a duration table, wherein the duration table includes a plurality of demand similarity interval values ​​and a first duration corresponding to each demand similarity interval value; Obtain the target demand similarity interval value according to the maximum demand similarity value; Obtain the corresponding first duration from the duration table according to the target demand similarity interval; Obtaining the start time of the first period according to the first duration and the end time of the first period; The first time period is obtained according to the start time of the first time period and the end time of the first time period.

6. The method for predicting customer demand based on machine learning according to claim 1, characterized in that: The steps of constructing a second time period, obtaining customer preference data in the second time period, and obtaining corresponding preference information according to the customer preference data include: Construct the second period; Obtain customer preference data in the second period; Acquire corresponding multiple customer preference items according to the customer preference data; Obtain multiple customer preference vectors corresponding to each customer preference item; A customer preference coefficient is obtained according to a plurality of customer preference items and a plurality of corresponding customer preference vectors, and is marked as preference information.

7. The method for predicting customer demand based on machine learning according to claim 6, characterized in that: The steps to construct the second period include: Obtaining the time point marked as deviant behavior information and marking it as the end time of the second period; Get the first duration of the first time period; Obtain standard customer preference coefficients; Obtain a second duration according to a customer preference coefficient, a standard customer preference coefficient, and a first duration; Obtaining the start time of the second period according to the second duration and the end time of the second period; The second time period is obtained according to the start time of the second time period and the end time of the second time period.

8. The method for predicting customer demand based on machine learning according to claim 1, characterized in that: The steps of obtaining customer demand behavior information based on the deviation behavior information, demand trend information and preference information, and obtaining customer demand information based on the customer demand behavior information include: Obtaining the corresponding current behavior feature vector according to the deviation behavior information; Obtain the corresponding standard behavior feature vector based on customer behavior data; Obtain the customer demand value based on the current behavior characteristic vector, demand trend information, and preference information, and mark it as customer demand behavior information; Obtaining a demand forecast table, wherein the demand forecast table includes a plurality of customer demand interval values ​​and customer demand information corresponding to each demand interval value; Obtain the corresponding target customer demand interval value according to the customer demand value; Obtain the corresponding customer demand information from the demand forecast table based on the target customer demand range value.

9. A customer demand forecasting system based on machine learning, applied to the customer demand forecasting method based on machine learning according to any one of claims 1 to 8, characterized in that: include: The behavior characteristics module is used to obtain customer behavior data and obtain current behavior characteristics information based on the behavior data; A behavior judgment module is used to obtain historical demand information of customers, obtain demand similarity information based on current behavior characteristic information and historical demand information of customers, and judge whether the demand similarity information meets the first preset condition. If not, it is marked as deviation behavior information; The first time period module is used to construct the first time period and obtain customer demand trend information within the first time period; The second time period module is used to construct the second time period, obtain the customer preference data in the second time period, and obtain the corresponding preference information according to the customer preference data; The demand matching module is used to obtain customer demand behavior information based on deviation behavior information, demand trend information and preference information, and obtain customer demand information based on customer demand behavior information.

10. A customer demand prediction terminal based on machine learning, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When one or more programs are executed by one or more processors, the one or more processors implement the customer demand prediction method based on machine learning as described in any one of claims 1 to 8.

Images