Consumption forecasting system and consumption forecasting method

Abstract

A consumption prediction system includes a storage device and a processor. The storage device is used to store historical environmental data and a historical consumption record. The processor is used to calculate a personal preference correlation coefficient, and input the historical environmental data, the historical consumption record and the personal preference correlation coefficient into a first type of neural network model, the first type of neural network model is used to generate a training model, when an accuracy rate of the training model is higher than a training threshold, the training model is regarded as a prediction model, a prediction environmental data is input into the prediction model, and the prediction model outputs a prediction result according to the prediction environmental data.

Description

Technical Field

[0001] The present invention relates to a forecasting system, and in particular to a consumption forecasting system and a consumption forecasting method. Background Art

[0002] Generally speaking, sales report analysis can only present past consumption records, because the number of store visitors and daily consumption preferences are difficult to accurately predict. The current number of store visitors may be estimated using traditional mathematical analysis or empirical rules. The prediction accuracy of these methods is low, resulting in the store's inability to accurately grasp the daily material preparation, which easily leads to insufficient material preparation or excessive material preparation, resulting in material waste. In addition, even if a variety of historical data is obtained, if human statistics or traditional mathematical analysis are used, a large amount of calculation is required, and it is difficult to quickly provide forecast data.

[0003] Therefore, how to effectively consider a variety of factors that affect consumption in order to quickly and accurately predict the number of store visitors and product sales has become one of the problems that need to be solved in this field. Summary of the invention

[0004] In order to solve the above problems, an aspect of the present disclosure provides a consumption prediction system, comprising a storage device and a processor. The storage device is used to store historical environmental data and a historical consumption record. The processor is used to calculate a personal preference correlation coefficient, and input historical environmental data, historical consumption records and personal preference correlation coefficients into a first type of neural network model. The first type of neural network model is used to generate a training model. When an accuracy rate of the training model is higher than a training threshold, the training model is regarded as a prediction model. A prediction environmental data is input into the prediction model. The prediction model outputs a prediction result based on the prediction environmental data, and obtains a predicted number of visitors and a predicted turnover in the prediction result. The predicted number of visitors is compared with an actual number of visitors to generate a correction weight for the number of visitors, and the predicted turnover is compared with an actual turnover to generate a correction weight for the turnover. It is determined whether the correction weight for the number of visitors is higher than a threshold value for the number of visitors weight or whether the correction weight for the turnover is higher than a threshold value for the turnover weight. When the correction weight for the number of visitors is higher than the threshold value for the number of visitors weight or the correction weight for the turnover is higher than the threshold value for the turnover weight, the training model is adjusted based on the correction weight for the number of visitors and the correction weight for the turnover to update the training model.

[0005] In order to solve the above-mentioned problems, another aspect of the present disclosure provides a consumption prediction method, comprising the following steps: calculating a personal preference correlation coefficient; inputting a historical environment data, a historical consumption record and a personal preference correlation coefficient into a first type of neural network model, the first type of neural network model is used to generate a training model; judging whether an accuracy of the training model is higher than a training threshold, when the accuracy of the training model is higher than the training threshold, the training model is regarded as a prediction model; inputting a prediction environment data into the prediction model, the prediction model outputs a prediction result based on the prediction environment data; obtaining a predicted number of visitors and a predicted turnover in the prediction result, comparing the predicted number of visitors with an actual number of visitors to generate a corrected weight for the number of visitors, and comparing the predicted turnover with an actual turnover to generate a corrected weight for the turnover; and judging whether the corrected weight for the number of visitors is higher than a threshold value for the number of visitors weight or whether the corrected weight for the turnover is higher than a threshold value for the turnover weight; when the corrected weight for the number of visitors is higher than the threshold value for the number of visitors weight or the corrected weight for the turnover is higher than the threshold value for the turnover weight, adjusting the training model according to the corrected weight for the number of visitors and the corrected weight for the turnover to update the training model.

[0006] As can be seen from the above, in the era of artificial intelligence, the amount of data is huge and complex. If you want to accurately and efficiently predict data, the factors and data volume to be considered are huge and complex. Therefore, data collection can no longer rely on manual maintenance. The consumption prediction system and consumption prediction method shown in the present invention analyze a large amount of complex data through a neural network model, and then generate a prediction model. The prediction model can output the predicted number of visitors and predicted turnover in the future. The predicted number of visitors can also be compared with the actual number of visitors to obtain the correction weight of the number of visitors, and the predicted turnover can be compared with the actual turnover to obtain the correction weight of the turnover. The prediction model can be adjusted through the correction weight of the number of visitors and the correction weight of the turnover to make the prediction model more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1A to Figure 1B FIG. 4 is a block diagram of a consumption forecasting system according to an embodiment of the present invention.

[0008] Figure 2 The present invention is a flowchart of a consumption forecasting method according to an embodiment of the present invention.

[0009] Figure 3A to Figure 3C is a schematic diagram illustrating a method for collecting environmental data according to an embodiment of the present invention.

[0010] Figure 4 The figure is a flow chart of a method for classifying consumer groups according to an embodiment of the present invention. DETAILED DESCRIPTION

[0011] The following description is a preferred implementation of the invention, and its purpose is to describe the basic spirit of the invention, but it is not intended to limit the invention. The actual content of the invention must refer to the scope of the following claims.

[0012] It must be understood that the words "comprise", "include" and the like used in this specification are used to indicate the existence of specific technical features, values, method steps, operation processing, elements and / or components, but do not exclude the addition of more technical features, values, method steps, operation processing, elements, components, or any combination of the above.

[0013] The words "first", "second", "third", etc. used in the claims are used to modify the elements in the claims. They are not used to indicate a priority order, a prior relationship, or that one element precedes another, or a temporal order in performing method steps. They are only used to distinguish elements with the same name.

[0014] Please refer to Figure 1A , Figure 1B and Figure 2 , Figure 1A FIG. 1 is a block diagram of a consumption forecasting system 100 according to an embodiment of the present invention. Figure 1B FIG. 1 is a block diagram of a consumption forecasting system 150 according to an embodiment of the present invention. Figure 2 FIG. 2 is a flow chart of a consumption forecasting method 200 according to an embodiment of the present invention.

[0015] like Figure 1A As shown, the consumption forecasting system 100 is applicable to an electronic device, such as a computer, a mobile phone, a tablet or other devices with computing functions. The consumption forecasting system 100 includes a storage device 10 and a processor 20. In one embodiment, the storage device 10 can be implemented as a read-only memory, a flash memory, a floppy disk, a hard disk, an optical disk, a flash drive, a magnetic tape, a database accessible by a network, or a storage medium with the same function that can be easily thought of by a person skilled in the art. In one embodiment, the processor 20 can be implemented by an integrated circuit such as a micro controller, a microprocessor, a digital signal processor, an application specific integrated circuit (ASIC) or a logic circuit.

[0016] like Figure 1BAs shown, the consumption prediction system 150 is applicable to an electronic device, such as a computer, a mobile phone, a tablet or other devices with computing functions. The consumption prediction system 150 includes a storage device 10, an environmental data collection module 12, a consumption record module 14, a consumer group classification module 16 and a model training module 18. In one embodiment, the environmental data collection module 12, the consumption record module 14, the consumer group classification module 16 and the model training module 18 are each implemented in a hardware circuit and are each coupled to the processor 20. In one embodiment, the environmental data collection module 12, the consumption record module 14, the consumer group classification module 16 and the model training module 18 are located in the processor 20. In one embodiment, the environmental data collection module 12, the consumption record module 14, the consumer group classification module 16 and the model training module 18 can each be implemented in software and executed by the processor 20. In one embodiment, the environmental data collection module 12, the consumption record module 14, the consumption group classification module 16 and the model training module 18 can be implemented together or separately by an integrated circuit such as a micro controller, a microprocessor, a digital signal processor, an application specific integrated circuit (ASIC) or a logic circuit.

[0017] The consumption forecasting method 200 is described below. The consumption forecasting method 200 can be implemented by the consumption forecasting system 100 or 150.

[0018] In step 210, processor 20 calculates a personal preference correlation coefficient.

[0019] In one embodiment, the processor 20 adds up the number of times a user spends at a specific store within a time interval to obtain the total number of times a user spends at a specific store, and divides the total number of times a user spends at a specific store by the number of times the user spends at all stores within the time interval to calculate the personal preference correlation coefficient. In one embodiment, the personal preference correlation coefficient calculation formula is as follows:

[0020]

[0021] The symbol t represents a time interval, such as 7, 15, 30, 60 days, etc. The symbol n represents the number of purchases in all stores at each time point. The symbol e represents the store, product type, or weather conditions, high, medium, low, etc. It is a variable that can be substituted with different values ​​depending on the actual application. For example, it is a store code, and the symbol R et Represents the number of times a customer visits a store at a certain point in time. Symbol: PAR etis the personal preference correlation coefficient for shopping in this store.

[0022] In step 220, the processor 20 inputs the historical environmental data, the historical consumption records and the personal preference correlation coefficient into a first type of neural network model, and the first type of neural network model is used to generate a training model.

[0023] In one embodiment, the first type of neural network model can be implemented by a known convolutional neural network (CNN) model. However, those skilled in the art should understand that the present invention is not limited to the use of CNN models, and other neural network models that can achieve automatic classification can also be applied.

[0024] In one embodiment, the historical environmental data may be collected by the environmental data collection module 12 or the processor 20. In one embodiment, the environmental data collected by the environmental data collection module 12 or the processor 20 may be regarded as the historical environmental data. In one embodiment, the environmental data may be obtained by the processor 20 or the environmental data collection module 12. Figure 3A to Figure 3C , Figure 3A to Figure 3C FIG. 3 is a schematic diagram illustrating an environmental data collection method 300 according to an embodiment of the present invention. The environmental data collection method 300 may be executed by the environmental data collection module 12 or the processor 20 . Figure 3A to Figure 3C The environmental data collection method 30 shown is not limited to an execution order and can also be executed simultaneously.

[0025] In one embodiment, if Figure 3A As shown, the processor 20 receives an explicit data 30 (such as weather, store evaluation, store name), stores the explicit data 30 in a first queue Q1, and reads the explicit data 31 from the first queue Q1, and obtains a text message in the explicit data 31, the text message is related to a store type 32 and an external factor 33. The store type 32 includes store evaluation, store location, traffic information, satisfaction, daily product information and / or activity information, etc. The external factor 33 includes information such as weather, season and / or temperature. In one embodiment, the user can manually input the explicit data 30 into the consumption prediction system 100 (or 150). In one embodiment, the processor 20 can obtain the explicit data 30 from the network.

[0026] In one embodiment, if Figure 3BAs shown, the processor 20 receives a digital text data 34, stores the digital text data 34 in a second queue Q2, and reads the digital text data 34 from the second queue Q2, and classifies the digital text data 34 by a second type of neural network model to obtain a store discussion heat 35 corresponding to the digital text data 34. In one embodiment, the processor 20 can obtain the digital text data 34 from the Internet. The digital text data 34 is, for example, text data in Facebook, Google, Twitter, or email. For example, the processor 20 can search for a discussion thread and evaluation score of a certain store through Google's search engine, and the accumulation of these evaluation scores can be regarded as the store discussion heat 35.

[0027] In one embodiment, the second type of neural network model can be implemented by a known CNN model. However, those skilled in the art should understand that the present invention is not limited to the use of CNN models, and other neural network models that can achieve automatic classification can also be applied. More specifically, after the processor 20 reads the digital text data 34 from the second queue Q2, the second type of neural network model can be used to determine the correctness of the digital text data 34 (for example, exclude the text in the advertising link and only leave the text in the user evaluation link), and then use the second type of neural network model to obtain the text information, and then use the second type of neural network model to give positive and negative evaluation scores to the obtained text information. The second type of neural network model can also be used to classify the discussion store topics in the digital text data 34 to classify the store types (such as snack bars, convenience stores, restaurants, etc.), and the number of articles corresponding to each store type is regarded as the store discussion heat 35. The correlation between the store type theme and the product type 37 can also be determined (for example, in the theme of buffet restaurants (store type), there are one thousand articles, and none of them mention steak (product type 37), so the correlation between the buffet restaurant type and steak is low).

[0028] In one embodiment, if Figure 3C As shown, the processor 20 receives a picture data 36 and stores the picture data 36 in a third queue Q3. The processor 20 reads the picture data 36 from the third queue Q3 and classifies the picture data 36 by a third type of neural network model to obtain a product type 37 corresponding to the picture data 36. In one embodiment, the user can manually input the picture data 36 into the consumption prediction system 100 (or 150). In one embodiment, the processor 20 can obtain the picture data 36 from the network. In one embodiment, the product type 37 is, for example, vegetarian, pork main meal, coffee, fruit, etc.

[0029] In one embodiment, the third type of neural network model can be implemented by a known CNN model. However, those skilled in the art should understand that the present invention is not limited to the use of CNN models, and other neural network models that can achieve automatic classification can also be applied.

[0030] In one embodiment, the first queue Q1 , the second queue Q2 , and the third queue Q3 may be combined into one queue.

[0031] In one embodiment, the environmental data collection module 12 or the processor 20 collects the explicit data 31 and the text content of the social networking site (such as digital text data 34). In terms of collecting the explicit data 31, the environmental data collection module 12 or the processor 20 can periodically call the application programming interface (API) of the external system or government agency to obtain external factors 33, such as: the degree of good or bad weather, season, whether it is a working day or not. Based on this information, it is helpful for the processor 20 to predict the motivation of consumers to choose merchants for consumption every day. For ordinary office workers, except for special activities, the merchants they prefer to consume may be fixed, especially in a private cloud environment, where employees and merchants usually do not change much. The explicit data 31 includes the content of Table 1:

[0032] date weather Activity temperature 11 / 1 sunny no 27 11 / 2 cloudy day no 25 11 / 3 cloudy day Sale 26

[0033] Table 1

[0034] In one embodiment, the external data 31 includes the data of the store and the service attributes provided by the store, such as: main dishes, the store's location in the mall, the number of positive reviews of the store, supported payment devices, the number of negative reviews of the store, business hours, etc. These will affect the number of customers visiting the store, and some consumers will consider store reviews and convenience. In one embodiment, the external data 31 includes the content of Table 2:

[0035]

[0036] Table 2 In one embodiment, the external data 31 includes the content of Table 3:

[0037] Store Employee Number Distance from the store Nearby stores Location 7-11 95031327 200m bookstore South Building MOS 95031327 400m Vitality North Building

[0038] Table 3

[0039] Since the distance between consumers and stores will also affect individual consumption intentions, and the neighboring stores of this store may also affect individual consumption tendencies, the explicit data 31 may also include the content of Table 3.

[0040] In addition, the environmental data collection module 12 or the processor 20 will also judge the positive and negative evaluation of the digital text data 34 (such as text messages and reply information content on social networking sites). Unlike the external data 31, these texts do not have certain rules and the correct content cannot be obtained through text interception. These texts may contain some false information, and the amount of digital text data 34 is huge, which is difficult to judge manually one by one or one sentence at a time. Therefore, these texts are judged for the correctness of the data through artificial intelligence algorithms (such as the second type of neural network model), and then the text information is classified to calculate the positive and negative evaluation of the store, which type of store is more popular, and the correlation between the store type and the product type 37 (for example, the number of articles containing a specific product is divided by the total number of articles on a specific topic).

[0041] In addition, the image data 36 can help define popular product types 37 (such as coffee) and store types (such as convenience stores), etc.

[0042] In the era of artificial intelligence, due to the complexity of external environmental factors and the huge and messy amount of data, data collection can no longer rely on manual maintenance. Therefore, through the above-mentioned environmental data collection method 300, by applying artificial intelligence algorithms (such as the second type of neural network model) to analyze data, efficient data collection can be achieved.

[0043] In one embodiment, the historical consumption records can be obtained by the consumption record module 14 or the processor 20. In one embodiment, the consumption records collected by the consumption record module 14 or the processor 20 can be regarded as historical consumption records. The consumption record module 14 or the processor 20 counts the consumer's consumption amount, consumption frequency, and average customer unit price according to the consumption records, and makes a ranking by year, month, and day. The consumption record is a good tool for predicting the number of visitors to the store. When the average customer unit price is low, the number of visitors may increase. When the average customer unit price is high, the number of visitors will decrease, but the total turnover may maintain a certain level. In one embodiment, the historical consumption records include the contents of Table 3:

[0044]

[0045] Table 3

[0046] Product type 37 is another factor that determines personal preference. Vegetarians tend to eat healthy meals and vegetarian meals, while those who do not eat beef tend to choose pork and mutton as their main meals. The amount of each product type 37 can be summed up and ranked by year and month to calculate the ranking and monthly sales amount, or the number of monthly consumption days can be calculated based on the number of visits to the store. In one embodiment, the historical consumption record includes the content of Table 4:

[0047] Product Type type staff Monthly sales amount Monthly Ranking Monthly consumption days vegetarian diet Main Meal 95031327 0 4 0 Pork main course Main Meal 95031327 600 1 12 Beef main course Main Meal 95031327 100 3 2 Chicken main course Main Meal 95031327 400 2 8 coffee Side dishes 95031327 1100 1 22 fruit Side dishes 95031327 700 1 14

[0048] Table 4 In one embodiment, the historical consumption record includes the content of Table 5:

[0049]

[0050] Table 5

[0051] The historical consumption records may include background information of the personnel who make mobile payments, such as age, job title, work type, etc. For example, some employees work in shifts, some often work overtime, some take leave at a certain time, etc. The processor 20 may group consumers based on this data.

[0052] See also Figure 4 , Figure 4 FIG. 4 is a flow chart of a method 400 for classifying consumer groups according to an embodiment of the present invention.

[0053] In one embodiment, the store type 32 , the external factors 33 , the store discussion popularity 35 , and the product type 37 are included in an external indicator 39 .

[0054] In one embodiment, the processor 20 can obtain the store ranking and sales 41 (e.g., daily sales) through Table 3, obtain the product preference ranking and sales 42 (e.g., monthly sales) through the monthly ranking in Table 4, and obtain the job type 43 through the job title in Table 5.

[0055] In one embodiment, store ranking and sales 41 , product preference ranking and sales 42 , and job type 43 are included in a consumption index 44 .

[0056] In one embodiment, the consumer group classification module 16 or the processor 20 may record the explicit indicator 39 and the consumption indicator 44 in a fourth queue Q4.

[0057] In one embodiment, the training model is stored in the storage device 10. In one embodiment, the first queue Q1, the second queue Q2, the third queue Q3 and the fourth queue Q4 are stored in the storage device 10.

[0058] In one embodiment, the consumer group classification module 16 or the processor 20 calculates a group correlation coefficient 45 according to the explicit indicator 39 or the consumption indicator 44 , and classifies the consumer groups according to the group correlation coefficient 45 and the personal preference correlation coefficient 46 .

[0059] In one embodiment, the consumer group classification module 16 or the processor 20 can read the explicit indicator 39 and the consumption indicator 44 from the fourth queue Q4 and calculate the group preference correlation coefficient 45. The group preference correlation coefficient 45 is calculated as follows:

[0060]

[0061] The symbol t represents a time interval, such as 7, 15, 30, 60 days, etc. The symbol n represents the number of purchases at all stores at each time point. The symbol e represents the store, product type, or weather conditions, high, medium, low, etc. The symbol R et Represents the number of times a group spends money at a particular store at each point in time. Symbol GAR et is the preference coefficient for consumption in this group (i.e., the group preference correlation coefficient is 45).

[0062] The group preference correlation coefficient can be divided into store consumption preference, product type purchase preference, weather quality and discussion heat preference index, etc. Substituting the relevant values ​​into the group preference correlation coefficient formula, the correlation coefficient of the consumption index and the correlation coefficient of the explicit index can be calculated. Another use of the group preference correlation coefficient 45 is to modify the definition of group members, such as: those who are easily affected by external weather group factors can be classified into the weather group. If it is calculated that the reference group (pre-defined group) is not highly correlated with the weather-related group preference coefficient, for example, regardless of the weather, the number of bento sold in the store is within a certain range, then the group members should be modified according to the personal preference correlation coefficient. The personal preference correlation coefficient has been described in step 210, so it will not be repeated here.

[0063] After the consumer group classification module 16 or the processor 20 calculates the correlation coefficient between the individual and the reference group (personal preference correlation coefficient), it can determine the correlation coefficient between the individual and the reference group. If the personal preference correlation coefficient is high (for example, the individual prefers vegetarianism, when the monthly consumption amount of vegetarianism is substituted as a parameter into the personal preference correlation coefficient formula, the personal preference correlation coefficient will be higher than the value after the monthly consumption amount of mutton main meal is substituted into the personal preference correlation coefficient formula), it means that the individual has a high correlation with the reference group (for example, the vegetarian group). If there is a high correlation with other reference groups, it means that the individual should be reclassified into another group. The correlation coefficient index between the individual and the reference group can be obtained by the following formula:

[0064]

[0065] The symbol g represents the group type. Symbol PAR et Represents the individual's preference coefficient. Symbol GAR et is the preference coefficient of each reference group. The symbol t represents the time interval, such as 7, 15, 30, 60 days, etc., and the symbol e represents the store, product type, or weather, high, medium, low, etc. The symbol PR eg It is an indicator of the correlation coefficient between an individual and a reference group. The larger the value, the more relevant it is. This can be used to classify consumer groups from multiple perspectives.

[0066] In step 230 , the processor 20 determines whether an accuracy of the training model is higher than a training threshold.

[0067] When the accuracy of the training model is higher than the training threshold, the processor 20 regards the training model as a prediction model.

[0068] When the accuracy of the training model is not higher than the training threshold, the processor 20 executes step 220 to input the personal preference correlation coefficient, more historical environmental data and historical consumption records into the first type of neural network model.

[0069] The result of model training is to accurately predict the number of visitors to the store. In combination with historical environmental data, historical consumption records, consumption records and personal preference correlation coefficients, there will definitely be omissions and missing values ​​in the data. Therefore, the missing values ​​will be filled by the average value of the personal background data and the group attribute to which the individual belongs. In one embodiment, the processor 20 is also used to determine whether the personal preference correlation coefficient, environmental data or historical consumption records contain a missing value. When the processor 20 determines that the personal preference correlation coefficient, historical environmental data or historical consumption records contain a missing value (for example, the weather data of a certain day is not collected), the processor automatically fills in the missing value (for example, the average value is calculated by the weather data of two adjacent days).

[0070] In one embodiment, the number of consumption records of each person in a store may be uneven, and the processor 20 will balance the data volume and increase or decrease the data volume to ensure the accuracy of the model prediction.

[0071] In one embodiment, the processor 20 inputs a training data into the training model to obtain a training result, and compares the training result with a known result (such as historical environmental data or historical consumption records) for similarity to calculate the accuracy of the training model.

[0072] In step 240 , the processor 20 inputs a predicted environment data into the prediction model, and the prediction model outputs a prediction result according to the predicted environment data.

[0073] For example, processor 20 inputs predicted environmental data such as the weather being cloudy, a sale being held, the temperature being 26 degrees, the product type being pork main course, and the job title being engineer into the prediction model, and the prediction results output by the prediction model are the number of visitors being 50 (predicted number of visitors), the turnover being 8,000 yuan (predicted turnover), and other prediction items (such as predicted number of consumptions, predicted ranking for the month, etc.).

[0074] In one embodiment, the predicted environment data may be part of the data included in the explicit indicator 39 and / or the consumption indicator 44 .

[0075] In one embodiment, the predicted environmental data can be regarded as data or conditional constraints for predicting a future time period (e.g., a week). The predicted environmental data may include, for example, date, weather, activity, temperature, store, evaluation, current collection, whether it is employee card consumption, business hours, location, convenience, satisfaction, employee number or store location...etc., which are conditions used for prediction.

[0076] In step 250, the processor 20 obtains a predicted number of visitors and a predicted turnover in the prediction result, compares the predicted number of visitors with an actual number of visitors to generate a visitor number correction weight, and compares the predicted turnover with an actual turnover to generate a turnover correction weight.

[0077] For example, if the predicted result is 50 visitors (for example, the predicted number of visitors next Wednesday) and the turnover is 8,000 yuan (for example, the predicted turnover next Wednesday), the processor 20 will compare the predicted number of visitors 50 with the actual number of visitors 75 (for example, the number of visitors after the actual settlement on next Wednesday) (since 75 is 1.5 times of 50, the correction weight of the number of visitors is 1.5), and will compare the predicted turnover of 8,000 yuan with the actual turnover of 9,600 yuan (for example, the turnover after the actual settlement on next Wednesday) (since 9,600 is 1.2 times of 8,000, the correction weight of the turnover is 1.2).

[0078] In step 260, the processor 20 determines whether the modified weight of the number of visitors (eg, 1.5) is higher than a threshold value of the number of visitors weight (eg, 1.2) or whether the modified weight of the turnover (eg, 1.3) is higher than a threshold value of the turnover weight (eg, 1.3).

[0079] In one embodiment, when the visitor number correction weight is higher than the visitor number weight threshold or the turnover correction weight is higher than the turnover weight threshold, step 220 is entered to input the historical environmental data, historical consumption records and personal preference correlation coefficients into the first type of neural network model, and the training model is adjusted according to the visitor number correction weight and the turnover correction weight to update the training model.

[0080] In one embodiment, when the modified weight of the number of visitors is not higher than the weight threshold of the number of visitors and the modified weight of the turnover is not higher than the weight threshold of the turnover, step 270 is entered.

[0081] In this example, the processor 20 determines that the modified weight of the number of visitors (e.g., 1.5) is higher than the weight threshold of the number of visitors (e.g., 1.2), and therefore, the weight threshold of the number of visitors is sent back to step 220. In some examples, the processor 20 sends both the weight threshold of the number of visitors and the modified weight of the turnover to step 220.

[0082] In one embodiment, the processor 20 can transmit the consumption tendency data of the same customer group on the day and / or the environmental data collected on the day to step 220, and the first type of neural network model in step 220 can obtain more information to adjust the training model.

[0083] In step 270, the prediction model outputs the predicted number of visitors and the predicted sales.

[0084] In one embodiment, the processor 20 or the prediction model can multiply the predicted number of visitors (e.g., 50 people) by the correction weight for the number of visitors (e.g., 1.5) to obtain a corrected result for the number of visitors (e.g., 75 people) during the current prediction or the next prediction, and multiply the predicted turnover (e.g., 8,000 yuan) by the correction weight for the turnover (e.g., 1.2) to obtain a corrected result for the turnover (e.g., 9,600 yuan).

[0085] In one embodiment, in the process of generating model prediction, the best model cannot be found in one go. The consumption prediction system 100 will repeat steps 210 to 230 to find the top few with higher accuracy and the lowest error rate as prediction models, and use the prediction models as the basis to predict the probability of an individual visiting the store. The prediction model will finally derive the ranking of stores and products recommended to the individual, and based on this ranking data, predict the number of store visitors within a time interval (e.g., a week), and according to the daily business situation and the individual's choice, modify the weights so that the number of visitors and turnover within a week can be within the correct range. Such a modified result can be fed back to the artificial intelligence model (i.e., return to step 220) and added to the input data of the model training. In addition, the output of the prediction model can be multiplied by the weight to modify the predicted value of the number of store visitors.

[0086] As can be seen from the above, in the era of artificial intelligence, the amount of data is huge and complex. If you want to accurately and efficiently predict data, the factors and data volume to be considered are huge and complex. Therefore, data collection can no longer rely on manual maintenance. The consumption prediction system and consumption prediction method shown in the present invention analyze a large amount of complex data through a neural network model, and then generate a prediction model. The prediction model can output the predicted number of visitors and predicted turnover in the future. The predicted number of visitors can also be compared with the actual number of visitors to obtain the correction weight of the number of visitors, and the predicted turnover can be compared with the actual turnover to obtain the correction weight of the turnover. The prediction model can be adjusted through the correction weight of the number of visitors and the correction weight of the turnover to make the prediction model more accurate.

[0087] [Explanation of symbols]

[0088] 100,150: Consumption forecasting system

[0089] 10: Storage device

[0090] 20: Processor

[0091] 12: Environmental data collection module

[0092] 14: Consumption record module

[0093] 16: Consumer group classification module

[0094] 18: Model training module

[0095] 200: Consumption Forecasting Methods

[0096] 210~270: Steps

[0097] 31: Explicit Data

[0098] Q1: First Queue

[0099] 32: Store Type

[0100] 33: External factors

[0101] 34: Digital text data

[0102] Q2: The second queue

[0103] 35: Store discussion popularity

[0104] 36: Image data

[0105] Q3: The third queue

[0106] 37: Product Type

[0107] 39: External indicators

[0108] 41: Store ranking and sales

[0109] 42: Product Preference Ranking and Sales

[0110] 43: Type of work

[0111] 44: Consumption indicators

[0112] Q4: The fourth cohort

[0113] 45: Group correlation coefficient

[0114] 46: Personal preference correlation coefficient

Claims

1. A consumption forecasting system, comprising: A storage device for storing historical environmental data and historical consumption records; and A processor is used to calculate a personal preference correlation coefficient, and input the historical environment data, the historical consumption record and the personal preference correlation coefficient into a first type neural network model, the first type neural network model is used to generate a training model, when an accuracy rate of the training model is higher than a training threshold, the training model is regarded as a prediction model, a prediction environment data is input into the prediction model, the prediction model outputs a prediction result according to the prediction environment data, and obtains a predicted number of visitors and a predicted turnover in the prediction result, compares the predicted number of visitors with an actual number of visitors to generate a correction weight for the number of visitors, and compares the predicted turnover with an actual turnover to generate a correction weight for the turnover, and determines whether the correction weight for the number of visitors is higher than a threshold value for the number of visitors or whether the correction weight for the turnover is higher than a threshold value for the turnover, when the correction weight for the number of visitors is higher than the threshold value for the number of visitors or the correction weight for the turnover is higher than the threshold value for the turnover, the training model is adjusted according to the correction weight for the number of visitors and the correction weight for the turnover to update the training model, in, The predicted environmental data includes a part of the explicit indicator, and the processor is further used to obtain the explicit indicator, including: Receiving an external data, a digital text data and an image data; Acquire text information in the external data, the text information being related to a store type and an external factor, Classifying the digital text data by a second type of neural network model to obtain a store discussion heat corresponding to the digital text data, and Classifying the image data by a third type of neural network model to obtain a product type corresponding to the image data; Among them, the store type, the external factors, the store discussion popularity and the product type are included in the external indicators.

2. A consumption prediction system as described in claim 1, wherein the processor sums up the number of times a user consumes in a specific store within a time interval to obtain a total number of consumptions in the specific store, and divides the total number of consumptions in the specific store by the number of consumptions of the user in all stores within the time interval to calculate the personal preference correlation coefficient.

3. The consumption prediction system as described in claim 1, wherein the processor is also used to determine whether the personal preference correlation coefficient, the historical environmental data or the historical consumption record contains a missing value, and when the processor determines that the personal preference correlation coefficient, the historical environmental data or the historical consumption record contains the missing value, the processor automatically fills in the missing value.

4. The consumption forecasting system as claimed in claim 1, wherein the processor inputs a training data into the training model to obtain a training result, and compares the training result with a known result for similarity to calculate the accuracy of the training model.

5. The consumption forecasting system as described in claim 1, wherein the processor determines that when the visitor number correction weight is not higher than the visitor number weight threshold and the turnover correction weight is not higher than the turnover weight threshold, the predicted visitor number is multiplied by the visitor number correction weight to obtain a visitor number correction result, and the predicted turnover is multiplied by the turnover correction weight to obtain a turnover correction result.

6. The consumption prediction system as claimed in claim 1, wherein the processor receives an external data, a digital text data and an image data including: The processor receives an external data and stores the external data in a first queue. The processor receives a digital text data and stores the digital text data in a second queue. The processor receives a picture data and stores the picture data in a third queue.

7. The consumption forecasting system as claimed in claim 6, wherein the processor obtains the external indicator comprising: The processor reads the external data from the first queue and obtains the text information in the external data. The processor reads the digital text data from the second queue, classifies the digital text data by the second type of neural network model, and obtains the store discussion popularity corresponding to the digital text data. The processor reads the image data from the third queue, classifies the image data by the third type of neural network model, and obtains the product type corresponding to the image data.

8. The consumption prediction system of claim 7, wherein the processor receives a consumption index, the consumption index comprising a job type, a product preference ranking and sales, and a store ranking and sales; in, The processor stores the explicit indicator and the consumption indicator in a fourth queue, calculates a group correlation coefficient according to the explicit indicator or the consumption indicator, and classifies consumer groups according to the group correlation coefficient and the personal preference correlation coefficient.

9. A consumption forecasting method, comprising: Calculate a person's preference correlation coefficient; Inputting a historical environmental data, a historical consumption record and the personal preference correlation coefficient into a first type of neural network model, the first type of neural network model is used to generate a training model; Determine whether an accuracy rate of the training model is higher than a training threshold, and when the accuracy rate of the training model is higher than the training threshold, regard the training model as a prediction model; Inputting a prediction environment data into the prediction model, and the prediction model outputs a prediction result according to the prediction environment data; Obtaining a predicted number of visitors and a predicted turnover in the prediction result, comparing the predicted number of visitors with an actual number of visitors to generate a visitor number correction weight, and comparing the predicted turnover with an actual turnover to generate a turnover correction weight; as well as Determine whether the modified weight of the number of visitors is higher than a threshold value of the number of visitors weight or whether the modified weight of the turnover is higher than a threshold value of the turnover weight; When the visitor number correction weight is higher than the visitor number weight threshold or the turnover correction weight is higher than the turnover weight threshold, the training model is adjusted according to the visitor number correction weight and the turnover correction weight to update the training model. The predicted environmental data includes a part of the explicit indicator, and the method further includes obtaining the explicit indicator, including: receiving an external data, a digital text data and an image data; and Acquire text information in the external data, the text information being related to a store type and an external factor, Classifying the digital text data by a second type of neural network model to obtain a store discussion heat corresponding to the digital text data, and Classifying the image data by a third type of neural network model to obtain a product type corresponding to the image data; Among them, the store type, the external factors, the store discussion popularity and the product type are included in the external indicators.

10. The consumption forecasting method according to claim 9, further comprising: When it is determined that the visitor number correction weight is not higher than the visitor number weight threshold and the turnover correction weight is not higher than the turnover weight threshold, the predicted visitor number is multiplied by the visitor number correction weight to obtain a visitor number correction result, and the predicted turnover is multiplied by the turnover correction weight to obtain a turnover correction result.

Images