Group restaurant behavior data multi-dimensional acquisition and extraction device and method

Abstract

The invention provides a group meal restaurant behavior data multi-dimensional extraction device and a group meal restaurant behavior data multi-dimensional extraction method, and aims to solve the problems of lack of consumer dining information acquisition means, incapability of well protecting consumer privacy, uneven acquired data quality, data missing and serious errors in the group meal industry. According to the method, data dimension division, data acquisition and cleaning, descriptive statistics, correlation analysis and the like can be performed on dining behaviors and information through multi-dimensional information acquisition, a specific mathematical model and a statistical analysis method under the condition that personal information of diners is not acquired, so that multi-dimensional information extraction and analysis of restaurant operation are realized; potential laws behind data are mined, restaurants are helped to better understand customer behaviors and demands, and service quality and operation benefits are improved.

Description

Technical Field

[0001] The present invention relates to the field of multi-dimensional information collection and analysis of group dining restaurants, and in particular to a device and method for multi-dimensional collection and extraction of behavioral data of group dining restaurants. Background Art

[0002] Group dining services serve a wide range of clients, including schools, businesses, government agencies, and other organizations. These clients vary significantly in their dietary preferences, spending power, and meal times. Accurate customer information is crucial to meeting these diverse needs.

[0003] Currently, group dining restaurants primarily use a variety of methods to collect customer information. For identity verification, card swiping and QR code scanning are common. For consumer information collection, smart checkout systems can record detailed information such as customer spending amount, time of purchase, and ordered dishes. This strongly physical, bound information collection method not only raises concerns about data security and privacy, but also compromises the dining experience for diners. The legal and compliant collection and use of customer information is a pressing issue. Furthermore, data quality varies, with some data potentially missing, erroneous, or incomplete, impacting the accuracy and effectiveness of data analysis.

[0004] In response to the above-mentioned status quo and existing problems, it is very important to collect and analyze the personal information of diners in group meals without obtaining their personal information, so as to help restaurants better understand customer behavior and needs, provide more timely and thoughtful services to diners, and operate restaurants more efficiently. Summary of the Invention

[0005] The purpose of the present invention is to target the dining information of group dining restaurants, without collecting the personal information of diners, through multi-dimensional information collection and matching, and use specific mathematical models and statistical analysis methods to perform data dimension division, data collection and cleaning, descriptive statistics, correlation analysis, etc. on dining behavior and information. After information extraction and mathematical model analysis, timely information on core elements of the restaurant such as food waste, food popularity, route planning, healthy diet, waiter scheduling optimization, dining behavior prediction, and service capabilities is obtained.

[0006] The present invention provides a multi-dimensional collection and extraction device for group dining restaurant behavior data, and the technical solution adopted is as follows: Multiple mobile phone location collector sites are regularly distributed on the top of the restaurant. All of these sites form a mobile phone location collector array through communication lines. When diners enter the restaurant, the mobile phone location collector array locates the diners while the entrance area camera collects the diners' facial and body features. After feature recognition, the feature information is associated with the network card address of the mobile device obtained by the mobile phone location collector array. Since diners carry their mobile phones with them throughout the restaurant, the above-mentioned device and method constitute a preliminary association module for diners, which can associate the diners' physical feature information and location information without obtaining the diners' identity information. The same method is used to collect the location information and change trajectory of restaurant waiters and identify their skills.

[0007] The indoor location collector array can infer the distance between each site and the signal transmitter based on the address of the mobile device and the received signal strength, and thus obtain the location of the diner in the signal array.

[0008] When the meal pickup mode is a buffet mode, a first type of tableware and meal information binding module is provided in the buffet area; the first tableware and meal information binding module includes: a tableware pickup table, an electronic tag tray, an electronic tag plate, a tray area camera, a tableware electronic tag reader / writer, a weighable buffet stove, a dish electronic tag reader / writer, and a meal pickup exit camera; the tableware pickup table is set at the meal pickup entrance of the buffet, and a tableware electronic tag reader / writer is set on the tabletop of the tableware pickup table, which can read the electronic tag tray and electronic tag plate obtained by the diners. The tray area camera is set at the buffet entrance and can clearly capture the position of the diner's image. Multiple weighable buffet stoves for placing dishes are placed in the meal pickup area and weigh the dishes in real time. The dish electronic tag reader / writer is placed on the side of the weighable buffet stove close to the diners, which is used to read the dishes taken from each electronic tag tray. The meal pickup exit camera is set at the exit of the meal pickup area, which is used to collect image information of the diners and all the dishes they have selected.

[0009] The working process of the first type of tableware and meal information binding module is as follows: after the diner gets the electronic tag tray and a certain number of electronic tag plates, the tableware electronic tag reader obtains the diner's electronic tag tray and electronic tag plate information, and then collects the multiple plate information and tray information sensed at the same time. The tray area camera installed in the buffet area collects the diner's image information and compares and matches it with the diner information collected by the door area camera. Combined with the address of the mobile device obtained by the mobile phone location collector array at this time, after a successful match, the one-to-one corresponding diner image information, location information, and plate information are obtained; during the diner's meal collection process, the tray information read by the dish electronic tag reader is matched with the dish reduction information obtained by the weighing buffet furnace to obtain the diner's meal quantity of the corresponding dish; the diner's location information from entering the restaurant to the end of picking up the meal is extracted and connected to obtain the diner's pre-meal movement line; the reduced number of each dish is extracted to obtain the consumption popularity of the corresponding dish.

[0010] When the meal pickup mode is the sales mode, a second type of tableware and meal information binding module is provided at the sales window; the second tableware and meal information binding module includes: an electronic tag tray, an electronic tag plate, a queuing area camera, and an electronic tag reader; the seller provides the electronic tag tray and the electronic tag plate to the diner, and a sales area electronic tag reader is provided on the countertop of the sales window to read the electronic tag tray and the electronic tag plate obtained by the diner. The queuing area camera is provided at the sales window to clearly capture the position of the image of the diners in the queue, the address of the mobile device obtained by the position collector array, the image information of the diner and the image information of the diners captured by the door area camera The seller compares and matches the information of the diner, and combines it with the address of the mobile device obtained by the location collector array at this time. After a successful match, the image information, location information, and plate information of the diner in the corresponding sales area are obtained; when the seller delivers the dishes to the diner, the seller places the dishes within the sensing range of the electronic tag reader in the sales area for reading, and obtains the type and number of dishes of the diner. Since the sales system sells in units of portions, the amount of food taken by the diner for the corresponding dishes is calculated; the location information of the diner from entering the restaurant to the end of picking up the food is extracted and connected to obtain the diner's pre-meal movement line; the number of dishes taken is extracted to obtain the consumption popularity of the dish.

[0011] The dining area camera is a public area security camera. The image information recorded in the dining area is specially stored. The storage time should be guaranteed not to be overwritten within one month. It is used to trace table seasoning poisoning and dining safety accidents. The mobile phone location collector array collects diners' position change information in the dining area and the time they stay at a certain location, and obtains diners' movement route information and dining time information after they go to the restaurant.

[0012] The catering waste information collection module includes: a leftovers collection table, a tray partition board, a leftovers collection area camera, a weighing sensor, and a leftovers collection electronic tag reader-writer; the tray partition board is composed of vertical boards that are continuously spaced apart, and the distance between two adjacent vertical boards in the tray partition board is greater than the side length of the tray. The tray partition board is fixed to the leftovers collection table, and a weighing sensor and a leftovers collection electronic tag reader-writer are set between every two vertical boards of the tray partition board and fixed on the leftovers collection table. A leftovers collection area camera is set above the leftovers collection table to collect images of diners who come to place leftovers after dining and the corresponding remaining dishes. After dining, diners come to the leftovers collection table to place their leftover electronic tag trays and electronic tag plates. The tray partition board is used to constrain the electronic tag tray to be placed above the weighing sensor and ensure that the leftovers collection electronic tag reader-writer can easily read the electronic tag tray information. The leftovers collection area camera collects images of diners and images of remaining dishes and matches them with the aforementioned diner preliminary association module and tableware and meal information binding module to form a closed loop of dining information.

[0013] In addition to the above-mentioned data collected in the restaurant, the data processing and analysis process also includes the dish data provided by the kitchen (including but not limited to the taste, smell, and texture of the dishes) and external environment data (including but not limited to date (day of the week, month, holiday), weather (temperature, precipitation, wind speed), and holiday activities).

[0014] A method for multi-dimensional collection and extraction of group dining restaurant behavior data adopts a technical solution, in which the obtained data processing flow includes: dividing all data into the following dimensions; converting the information with divided dimensions into mathematical models to define a multi-dimensional data space; data cleaning; slicing, dicing, drilling, rotating, and aggregating the formed data cubes; and extracting and analyzing the multi-dimensional information in the following dimensions through mathematical models.

[0015] All data are divided into the following dimensions: (1) User characteristics include: age, gender, body shape, dining time, etc.; (2) Dish attribute dimensions include: dish name, ingredient composition (protein / carbohydrate / fat content), cooking method, taste label (spicy / sweet / salty), dish classification (staple food / meat dish / vegetarian dish / soup), etc.; (3) Leftover meal data dimensions include: remaining weight, remaining dish type, remaining dish category (cold dish or hot dish), remaining time (lunch / after dinner), etc.; (4) External environment dimensions include: season, weather temperature, holiday signs, etc.; (5) Restaurant environment dimensions include: restaurant temperature, cafeteria seating density, waiting time in line, etc.; (6) Transaction data dimensions include: dish price, set meal combination, consumption amount, etc.

[0016] The information divided into dimensions is transformed into a mathematical model to define a multidimensional data space; the multidimensional data space is assumed to be composed of multiple dimension sets; each dimension has a hierarchical structure, for example, the hierarchy of the time dimension can be year → quarter → month; the measurement is a numerical attribute (such as the sales volume of dishes, the number of diners); a fact table is formed to represent the combination relationship between dimensions and measurements, and each record corresponds to a fact.

[0017] Data cleaning: Missing value processing, such as using interpolation for missing fields of user features; outlier detection, such as isolating and analyzing records where the amount of leftover food exceeds 200% of the initial amount; time series alignment, such as synchronizing transaction timestamps with environmental data to minute-level accuracy.

[0018] Slice, dice, drill, rotate, and aggregate the formed data cube.

[0019] The following dimensions are extracted and analyzed from multi-dimensional information through mathematical models: (1) Attribution analysis of food waste The attribution analysis of food waste is carried out by collecting information on diners' meal pick-up, leftovers, and the consumption of each dish. The waste intensity of each dish is obtained by slicing and extracting from the dimension of a single dish, and the relative waste index of each dish is obtained by slicing the waste intensity of all dishes. Combined with the restaurant's environmental information and the taste and flavor information of the dishes in the kitchen, through data aggregation processing, a correlation map between sensory perception (A too salty, B taste discomfort, C greasy, D early satiety, E difficulty chewing, F increased dining time) and waste (waste intensity, relative waste index) is established.

[0020] (2) Attribution analysis of dish popularity Dish popularity attribution analysis collects information about diners picking up their meals and the consumption of each dish, segments all dishes into time dimensions, and establishes the influence of dish popularity over time (for example, the popularity of newly launched dishes will gradually decrease) and other factors (such as user reviews, sales, seasonal changes, etc.). By combining these factors, a more complex dynamic model of dish popularity is constructed to predict and manage the changes in dish popularity over different time periods.

[0021] (3) Restaurant traffic flow optimization Restaurant traffic flow optimization is based on the collected location information of diners and their changes. The restaurant's location nodes represent functional areas (such as entrances, dining tables, food pickup areas, cash registers, etc.), edges represent feasible paths, and weights are the movement costs (time or distance). The goal is to achieve the shortest average movement path for diners by optimizing node layout and path weights; maximize the separation between waiter and diner paths; and minimize traffic bottlenecks at key nodes.

[0022] (4) Healthy diet assessment Based on the food supply data, the nutritional components of each dish (calories, protein, fat, carbohydrates, sodium, dietary fiber, vitamins, etc.) are standardized, and the cooking methods (steaming, stir-frying, deep-frying, etc.) and ingredient composition (proportion of vegetables, type of meat, etc.) are marked. Combined with the collected diner data: selected dishes and portions (weighed by the card swiping system or smart plate), feature profiling is performed: age, gender, BMI, history of chronic diseases (hypertension / diabetes, etc.), job type; feature derivation: individual health indicators such as the deviation between daily nutrient intake and recommended daily intake (DRIs), dietary pattern characteristics: breakfast regularity, vegetable diversity index, and the ratio of coarse and fine grains in staple foods; environmental correlation characteristics: the correlation coefficient between temperature and calorie intake.

[0023] (5) Waiter scheduling optimization Based on the collected waiter location information and its changing patterns, the core variables are defined by comparing the number, location, waiting time, and dining duration information of diners. The business hours (e.g., 8:00-24:00) are divided into T time periods (e.g., each 30-minute period). The waiter set: S = s1, s2, ..., sN}, including full-time / part-time employees, skill labels (e.g., service type) and service demand: Dt = the number of customers to be served in the tth period. A decision is made on whether waiter s is on duty in the tth period.

[0024] (6) Dining behavior prediction The collected structured data, such as the number of diners per day / per period, sales of each dish, and repurchase rate, are normalized; external data such as date and weather are subjected to time feature extraction, and the date is converted into features such as the day of the week, month, and whether it is a holiday; missing data are filled using interpolation or mean.

[0025] The total customer flow is predicted using time series plus external variables as variables; a regression model of sales volume, total number of customers, and time characteristics is established for each dish to predict dish demand; a user-dish matrix is constructed, and consumer preferences are predicted by predicting unobserved consumer preferences through matrix decomposition.

[0026] (7) Service capability analysis The collected data (number of diners in period t, number of employees in the jth category (chefs, waiters, etc.), average dining time of diners, consumption of dishes in the kth category, discarded amount of dishes in the kth category (leftovers + discarded by customers), working hours of employees in the jth category) are used as core input variables, and the service capability score in period t, work efficiency of employees in the jth category, and resource waste index are used as output variables to achieve: maximizing customer service efficiency (reducing waiting time), minimizing resource waste (amount of leftovers, employee idle rate), and balancing workload (avoiding excessive fatigue of employees).

[0027] Compared with the prior art, the present invention has the following beneficial effects: 1. Existing technologies for collecting customer information often rely on card swiping, code scanning (QR codes, barcodes), and facial recognition to record customer identity information and dining times. Smart checkout systems also record customer spending amounts, time of purchase, and ordered dishes. These collection methods pose privacy risks and can also generate negative feedback.

[0028] The present invention adopts a method of preliminary matching of the position acquisition array with the diner's image features, and further matching of the position acquisition array and the diner's image feature electronic tag tray. This secondary dimensional matching device and method associates the diner's physical feature information and position information and collects data on the dining situation without obtaining the diner's identity information. On the one hand, this ensures the diner's privacy, and on the other hand, increases the accuracy of data matching, which is beneficial to the reduction of analysis accuracy caused by data missing or anomalies in the subsequent data processing process.

[0029] 2. The device used in this invention achieves closed-loop collection of restaurant dining data at a low cost. It segments data according to user characteristics, dish attributes, leftover data, external environment, restaurant environment, and transaction data. Combining data cleaning, descriptive statistics, and correlation analysis, this information extraction and analysis enables analysis of dish waste, dish popularity, route planning, healthy eating, waiter scheduling optimization, dining behavior prediction, and service capacity. This solution achieves a closed loop from data collection to intelligent decision-making through multi-dimensional sensor data fusion, spatiotemporal feature engineering, and hybrid modeling technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is the overall schematic diagram of the restaurant and installation; Figure 2 It is a schematic diagram of the diner’s preliminary association module; Figure 3 This is a schematic diagram of the first type of tableware and meal information binding module; Figure 4 This is a schematic diagram of the second type of tableware and meal information binding module; Figure 5 This is a schematic diagram of the food waste information collection module. DETAILED DESCRIPTION

[0031] like Figure 1-5As shown, multiple mobile phone location collector sites 011 are regularly distributed on the top of the restaurant. All mobile phone location collector sites 011 form a mobile phone location collector array 012 through communication lines. When diners enter the restaurant, the mobile phone location collector array 012 locates the diners while the door area camera 014 collects the diner's facial features and body features. After feature recognition, the feature information is associated with the network card address of the mobile device obtained by the mobile phone location collector array 012. Since diners carry their mobile phones 013 with them throughout the restaurant, the above-mentioned device and method constitute a preliminary association module for diners, which can associate the diner's physical feature information and location information without obtaining the diner's identity information. The same method is used to collect the location information and change trajectory of the restaurant waiters and identify their skills.

[0032] Indoor location collection methods include Wi-Fi positioning, Bluetooth positioning, ultra-wideband positioning, and hybrid positioning. Considering that most mobile phones may have Bluetooth and NFC turned on much less than wireless networks, this patent preferably uses Wi-Fi positioning. Since the WiFi network protocol stipulates that mobile devices accessing access points will exchange data frames when not connected, the data frames contain the mobile device's MAC (Media Access Control) address and RSSI (Received Signal Strength Indication) received signal strength. The strength of the signal value is roughly inversely proportional to the distance between the two. Using the RSSI signal strength value to infer the distance between each station in the WiFi collector array and the signal transmitter, the diner's position in the signal array can be obtained.

[0033] When the meal pickup mode is buffet mode, a first type of tableware and meal information binding module is provided at the buffet area; the first tableware and meal information binding module includes: a tableware pickup table 021, an electronic tag tray 022, an electronic tag meal plate 023, a tray area camera 024, a tableware electronic tag reader 025, a weighable buffet oven 026, a dish electronic tag reader 027, and a meal pickup exit camera 028; the tableware pickup table 021 is provided at the meal pickup entrance of the buffet, and a tableware electronic tag reader 025 is provided on the tabletop of the tableware pickup table 021, which can read The diners obtain an electronic tag tray 022 and an electronic tag meal plate 023. The tray area camera 024 is set at a position at the buffet entrance where the diner's image can be clearly captured. Multiple weighable buffet ovens 026 for placing dishes are placed in the meal pick-up area and weigh the dishes in real time. The dish electronic tag reader 027 is placed on the side of the weighable buffet oven 026 close to the diners to read the dishes taken from each electronic tag tray 022. The meal pick-up exit camera 028 is set at the exit of the meal pick-up area to collect image information of the diners and all the dishes they have selected.

[0034] Preferably, the electronic tags in the electronic tag tray 022, the electronic tag dinner tray 023, and the electronic tag reader / writer of this patent adopt wireless radio frequency identification technology (RFID), wherein the electronic tags in the electronic tag tray 022 and the electronic tag dinner tray 023 adopt passive RFID.

[0035] The working process of the first type of tableware meal information binding module is as follows: after the diner gets the electronic tag tray 022 and a certain number of electronic tag plates 023, the tableware electronic tag reader 025 obtains the diner's electronic tag tray 022 and electronic tag plate 023 information, and then collects the information of multiple plates and trays sensed at the same time. The tray area camera 024 set in the buffet area collects the diner's image information and compares and matches it with the diner information collected by the door area camera 014. Combined with the WiFi collector at this time, The MAC address of the mobile device obtained by the array is matched successfully to obtain the corresponding diner's image information, location information, and plate information; during the diner's meal collection process, the tray information read by the dish electronic tag reader 027 is matched with the dish reduction information obtained by the weighing buffet furnace 026 to obtain the diner's take-away quantity of the corresponding dish; the diner's location information from entering the restaurant to the end of picking up the meal is extracted and connected to obtain the diner's pre-meal movement line; the reduced number of each dish is extracted to obtain the consumption popularity of the corresponding dish.

[0036] When the meal pickup mode is the sales mode, a second type of tableware and meal information binding module is provided at the sales window; the second tableware and meal information binding module includes: an electronic tag tray 022, an electronic tag plate 023, a queuing area camera 031, and a sales area electronic tag reader 032; the seller provides the electronic tag tray 022 and the electronic tag plate 023 to the diner, and a sales area electronic tag reader 032 is provided on the countertop of the sales window for reading the electronic tag tray 022 and the electronic tag plate 023 obtained by the diner, and the queuing area camera 031 is provided at the sales window to clearly capture the position of the image of the diners in the queue, the MAC address of the mobile device obtained by the WiFi collector array, and the image information of the diner. The information is compared and matched with the diner information collected by the door area camera 014, and combined with the MAC address of the mobile device obtained by the WiFi collector array at this time, the corresponding diner image information, location information, and plate information of the sales area are obtained after a successful match; when the seller delivers the dishes to the diner, the seller places the dishes within the sensing range of the electronic tag reader 032 in the sales area for reading, and obtains the type and number of dishes of the diner. Since the sales system sells in units of portions, the amount of food taken by the diner for the corresponding dishes is calculated; the location information of the diner from entering the restaurant to the end of picking up the food is extracted and connected to obtain the diner's pre-meal movement line; the number of dishes taken is extracted to obtain the consumption popularity of the dish.

[0037] The camera in the dining area is the public area security camera 033. The image information recorded in the dining area is specially stored. The storage time should be guaranteed not to be overwritten within one month. It is used to trace the poisoning of table seasonings and dining safety accidents. The WiFi collector array collects the location change information of diners in the dining area and the time they stay in a certain location, and obtains the movement route information and dining time information of diners after they go to the restaurant.

[0038] The catering waste information collection module includes: a leftover collection table 041, a tray partition board 042, a leftover collection area camera 043, a weighing sensor 044, and a leftover collection electronic tag reader 045; the tray partition board 042 is composed of vertical plates distributed at continuous intervals, and the distance between two adjacent vertical plates in the tray partition board is greater than the side length of the tray. The tray partition board 042 is fixed on the leftover collection table 041, and a weighing sensor 044 and a leftover collection electronic tag reader 045 are set between every two vertical plates of the tray partition board 042 and fixed on the leftover collection table 041. A leftover collection area camera 043 is set above the leftover collection table 041 to collect images of diners who come to place leftover food after dining and the corresponding remaining images of dishes. After dining, diners come to the leftovers collection counter 041 to place their leftover electronic tag tray 022 and electronic tag meal plate 023. The tray partition plate 042 is used to constrain the electronic tag tray 022 to be placed above the weighing sensor 044 and ensure that the leftovers collection electronic tag reader 045 can easily read the information of the electronic tag tray 022. The leftovers collection area camera 043 collects images of diners and images of remaining dishes and matches them with the aforementioned diner preliminary association module and tableware and meal information binding module to form a closed loop of dining information.

[0039] When the scrap collection platform 041 is manually collected, preferably, the pallet partition plate 042 is fixed on the table top of the scrap collection platform 041; when the scrap collection platform 041 is automatically collected by a conveyor belt, preferably, the pallet partition plate 042 is fixed on the table top of the scrap collection platform 041 and extends above the conveyor belt.

[0040] In addition to the above-mentioned data collected in the restaurant, the data processing and analysis process also includes the dish data provided by the kitchen (including but not limited to the taste, smell, and texture of the dishes) and external environment data (including but not limited to date (day of the week, month, holiday), weather (temperature, precipitation, wind speed), and holiday activities).

[0041] A multi-dimensional collection and extraction method for group dining restaurant behavior data includes the following data processing procedures: dividing all data into the following dimensions; converting the divided dimension information into mathematical models to define a multi-dimensional data space; data cleaning; slicing, dicing, drilling, rotating, and aggregating the formed data cubes; and extracting and analyzing the following dimensions of the multi-dimensional information through mathematical models.

[0042] The following describes the data processing process obtained in conjunction with a specific embodiment. All data are divided into the following dimensions: (1) User characteristics include: age, gender, body shape, dining time, etc.; (2) Dish attribute dimensions include: dish name, ingredient composition (protein / carbohydrate / fat content), cooking method, taste label (spicy / sweet / salty), dish classification (staple food / meat dish / vegetarian dish / soup), etc.; (3) Leftover meal data dimensions include: remaining weight, remaining dish type, remaining dish category (cold dish or hot dish), remaining time (lunch / after dinner), etc.; (4) External environment dimensions include: season, weather temperature, holiday signs, etc.; (5) Restaurant internal environment dimensions include: restaurant temperature, cafeteria seating density, queue waiting time, etc.; (6) Transaction data dimensions include: dish price, set meal combination, consumption amount, etc.

[0043] The dimensioned information is transformed into a mathematical model to define a multidimensional data space. Assume that the multidimensional data space consists of a set of dimensions (D = {D1, D2, … Di}). Each dimension Di has a hierarchical structure Li = {Li1}, Li2…Lik}. For example, the hierarchy of the time dimension can be year → quarter → month. The measurement set: M = {m1, m2… mp}, each measurement is a numerical attribute (such as the sales volume of dishes, the number of diners). The fact table is formed: F subseteq D1 × D2 ×…×Dn × M, which represents the combination relationship between dimensions and measurements. Each record f∈F corresponds to a fact.

[0044] Data cleaning: Missing value processing, such as using the KNN interpolation method for missing fields of user features; outlier detection, such as isolating and analyzing records where the amount of leftover food exceeds 200% of the initial amount; time series alignment, such as synchronizing transaction timestamps with environmental data to minute-level accuracy.

[0045] Data cube and OLAP operations include: slicing, dicing, drilling, pivoting, and aggregation.

[0046] Slicing transforms a data cube into a lower-dimensional subset by selecting a value from a specific dimension. For example, when launching a new dish, selecting sales data for that dish over time from a dish sales data cube that includes time, dish, and region dimensions allows for focused analysis of the dish's sales performance during a specific time period, helping restaurant managers quickly filter out irrelevant information and focus on analyzing the dish's popularity.

[0047] Slicing forms a smaller sub-cube by selecting specific values from multiple dimensions. For example, in a sales data cube containing time, product, and region dimensions, data from a specific year and region can be selected. Slicing allows analysts to conduct in-depth analysis across multiple dimensions, providing more granular control over data and optimizing food raw material inventory and sales strategies.

[0048] Drilling operations come in two forms: drill-down and drill-up. Drilling down allows analysts to view more detailed data, while drilling up allows them to return to higher-level summary data. The advantage of drill-down operations lies in their flexibility and dynamism, allowing data analysts to freely switch between different levels of aggregation, thereby obtaining a more comprehensive analytical perspective. This operation is particularly important in dynamic data analysis and real-time decision-making.

[0049] Pivoting provides different perspectives by changing the arrangement of dimensions in a data cube. For example, swapping the time and product dimensions allows analysis of sales figures for different products over different time periods. The advantage of pivoting is that it allows analysts to examine data from multiple perspectives, uncovering hidden patterns and trends. Pivoting allows for a better understanding of the multidimensional nature of data, discovering relationships within the data, and conducting exploratory data analysis to uncover valuable insights.

[0050] Aggregation operations simplify data cubes by summarizing data. For example, daily sales data can be aggregated into monthly sales data, or product-level data can be aggregated into product category data. This operation can significantly improve data processing efficiency.

[0051] The dimensions and methods of extracting multi-dimensional information about restaurant dining include the following processes: (1) Attribution analysis of food waste The attribution analysis of food waste is carried out by collecting information on diners' meal pick-up, leftovers, and the consumption of each dish. The waste intensity of each dish is obtained by slicing and extracting from the dimension of a single dish, and the relative waste index of each dish is obtained by slicing the waste intensity of all dishes. Combined with the restaurant's environmental information and the taste and flavor information of the dishes in the kitchen, through data aggregation processing, a correlation map between sensory perception (A too salty, B taste discomfort, C greasy, D early satiety, E difficulty chewing, F increased dining time) and waste (waste intensity, relative waste index) is established.

[0052] Preferably, the core indicator definition of attribution analysis of food waste is selected as Waste intensity of dish d: Wd = Remaining quantity of dish d / Served quantity of dish d × 100% Relative waste index of each dish:

[0053] Preferably, a Bayesian hierarchical model is used to regress the data: Wd ∼Beta( μdϕ ,(1− μd ) ϕ ) logit( μd )= α + β 1 x 1 d +⋯+ bkxkd + cuisine + meal cuisine ∼N(0, σc 2 ) meal ∼N(0, sm 2 ) c : Random effect of dish category (such as meat / vegetarian / staple food) d : Meal period random effect (breakfast / lunch / dinner) (2) Attribution analysis of dish popularity Dish popularity attribution analysis collects information about diners picking up their meals and the consumption of each dish, segments all dishes into time dimensions, and establishes the influence of dish popularity over time (for example, the popularity of newly launched dishes will gradually decrease) and other factors (such as user reviews, sales, seasonal changes, etc.). By combining these factors, a more complex dynamic model of dish popularity is constructed to predict and manage the changes in dish popularity over different time periods.

[0054] The preferred model is trained and validated based on real data based on Newton's law of cooling to ensure its accuracy and practicality.

[0055] (3) Restaurant traffic flow optimization Restaurant traffic flow optimization is based on the collected location information of diners and their changes. The restaurant's location nodes represent functional areas (such as entrances, dining tables, food pickup areas, cash registers, etc.), edges represent feasible paths, and weights are the movement costs (time or distance). The goal is to achieve the shortest average movement path for diners by optimizing node layout and path weights; maximize the separation between waiter and diner paths; and minimize traffic bottlenecks at key nodes.

[0056] Preferably, the movement line optimization is performed using the model that minimizes the total movement cost:

[0057] Among them, the area set: V={v1,v2,…,vn}, represents the n functional areas of the restaurant; the edge set: E={(vi,vj)|areas vi and vj are directly accessible}, and the weight wij is the movement cost; the flow matrix: F=[fij], represents the unit time flow from area vi to vj (such as the number of customer / waiter moves); the layout variable: xik∈{0,1}, is 1 if area vi is assigned to position k (discretized spatial coordinates), otherwise it is 0.

[0058] (4) Healthy diet assessment Based on the food supply data, the nutritional components of each dish (calories, protein, fat, carbohydrates, sodium, dietary fiber, vitamins, etc.) are standardized, and the cooking methods (steaming, stir-frying, deep-frying, etc.) and ingredient composition (proportion of vegetables, type of meat, etc.) are marked. Combined with the collected diner data: selected dishes and portions (weighed by the card swiping system or smart plate), feature profiling is performed: age, gender, BMI, history of chronic diseases (hypertension / diabetes, etc.), job type; feature derivation: individual health indicators such as the deviation between daily nutrient intake and recommended daily intake (DRIs), dietary pattern characteristics: breakfast regularity, vegetable diversity index, and the ratio of coarse and fine grains in staple foods; environmental correlation characteristics: the correlation coefficient between temperature and calorie intake.

[0059] (5) Waiter scheduling optimization Based on the collected waiter location information and its changing patterns, the core variables are defined by comparing the number, location, waiting time, and dining duration information of diners. The business hours (e.g., 8:00-24:00) are divided into \( T \) time periods (e.g., each 30 minutes is a time period); the waiter set: S = s1, s2, ..., sN}, including full-time / part-time employees, skill labels (e.g., service type) and service demand: Dt = the number of customers to be served in the tth time period, and a decision is made on whether waiter s is on duty in the tth time period.

[0060] Preferably, the decision is made by minimizing the total cost plus penalty model:

[0061] Where cs is the time period wage cost of waiter s (including overtime premium); λ is the penalty coefficient for unmet demand (reflecting the weight of service quality).

[0062] (6) Dining behavior prediction The collected structured data, such as the number of diners per day / per period, sales of each dish, and repurchase rate, are normalized; external data such as date and weather are subjected to time feature extraction, and the date is converted into features such as the day of the week, month, and whether it is a holiday; missing data are filled using interpolation or mean.

[0063] The total customer flow is predicted using time series plus external variables as variables; a regression model of sales volume, total number of customers, and time characteristics is established for each dish to predict dish demand; a user-dish matrix is constructed, and consumer preferences are predicted by predicting unobserved consumer preferences through matrix decomposition.

[0064] Preferably, the following mathematical models are used for total customer flow prediction, dish demand prediction, and consumer preference prediction: Polynomial model for total passenger flow prediction: y ( t )= g ( t )+ s ( t )+ h ( t )+ β ⋅ X ext( t )+ ϵt in, g ( t ) is the trend term (piecewise linear or logistic growth), s ( t ) is the seasonal term (Fourier series modeling weekly / annual cycles), h ( t ) is the holiday effect, X ext( t ) is an external variable (e.g., a weighted combination of promotions and weather).

[0065] Multiple regression model for dish demand prediction: Sales i ( t )= αi + βi ⋅Total_Customers( t )+ for ⋅Weekend( t )+ two ⋅Sales i ( t −1)+ ϵi ( t ) Among them, Weekend( t) is whether it is a weekend, and random forest / XGBoost can be used to handle nonlinear relationships.

[0066] Consumption preference model: R≈U·V Among them, U∈Rm×k is the user potential feature matrix, and V∈Rn×k is the dish potential feature matrix.

[0067] (7) Service capability analysis The collected data (number of diners in period t, number of employees in the jth category (chefs, waiters, etc.), average dining time of diners, consumption of dishes in the kth category, discarded amount of dishes in the kth category (leftovers + discarded by customers), working hours of employees in the jth category) are used as core input variables, and the service capability score in period t, work efficiency of employees in the jth category, and resource waste index are used as output variables to achieve: maximizing customer service efficiency (reducing waiting time), minimizing resource waste (amount of leftovers, employee idle rate), and balancing workload (avoiding excessive fatigue of employees).

[0068] Preferably, to quantify restaurant service capabilities and optimize resource allocation, the employee efficiency model uses the following mathematical model: Hey ( t )= αj ⋅[1− Hj ( t ) / H max] ⋅ [ Sj actual( t ) / Sj required( t )] in, αj Baseline efficiency for employee type (e.g. chef = 1.2, waiter = 1.0), H max is the maximum reasonable working time (e.g. 8 hours), Sj actual is the number of employees actually assigned, Sj Required is the theoretical number of employees required (calculated by the amount of tasks).

[0069] In summary, the present invention provides a multi-dimensional extraction device and method for group meal restaurant behavior data. Without obtaining the personal information of diners, it realizes multi-dimensional information collection through the setting of special information equipment. Combined with specific mathematical models and statistical analysis methods, it can realize the extraction and analysis of multi-dimensional information of restaurant operations, helping restaurants to better understand customer behavior and needs.

Claims

1. A multi-dimensional collection and extraction device for group dining restaurant behavior data, characterized in that: include: Multiple mobile phone location collector sites are regularly distributed on the top of the restaurant; all mobile phone location collector sites form a mobile phone location collector array through communication lines, and the location collector array infers the distance between each site and the signal transmitter based on the address of the mobile device and the strength of the received signal to obtain the location of the diner in the signal array; it also includes a door area camera set at the entrance of the restaurant. When the diner enters the restaurant, the mobile phone location collector array locates the diner while the door area camera collects the diner's facial features and body features, and after feature recognition, associates the feature information with the network card address of the mobile device obtained by the mobile phone location collector array; since the diners carry their mobile phones with them throughout the restaurant, the above-mentioned device and method constitute The preliminary association module for diners can associate the physical characteristics and location information of diners without obtaining their identity information; the same method is used to collect the location information and change trajectory of restaurant waiters and identify their skills; the buffet area is equipped with a first-class tableware and meal information binding module; the sales window is equipped with a second-class tableware and meal information binding module; there is also a food waste information collection module; the dining area is equipped with public area security cameras, which record image information of the dining area and store it specifically for tracing dining safety accidents including table seasoning poisoning; the mobile phone location collector array collects diners' location change information in the dining area and the time they stay at a certain location, and obtains the diners' movement route information and meal time information after they go to the restaurant.

2. The multi-dimensional collection and extraction device for group dining restaurant behavior data according to claim 1 is characterized in that: The first tableware and meal information binding module includes: a tableware collection table, an electronic tag tray, an electronic tag plate, a tray area camera, a tableware electronic tag reader / writer, a weighable buffet stove, a dish electronic tag reader / writer, and a meal pick-up exit camera; the tableware collection table is set at the meal pick-up entrance of the buffet, and a tableware electronic tag reader / writer is set on the table top of the tableware collection table, which can read the electronic tag tray and electronic tag plate obtained by the diners. The tray area camera is set at the entrance of the buffet and can clearly capture the position of the diner's image. Multiple weighable buffet stoves for placing dishes are placed in the meal pick-up area and weigh the dishes in real time. The dish electronic tag reader / writer is placed on the side of the weighable buffet stove close to the diners, which is used to read the dishes taken from each electronic tag tray. The meal pick-up exit camera is set at the exit of the meal pick-up area, which is used to collect image information of the diners and all selected dishes.

3. The multi-dimensional collection and extraction device for group dining restaurant behavior data according to claim 1 is characterized in that: The second tableware and meal information binding module includes: an electronic tag tray, an electronic tag meal plate, a queue area camera, and an electronic tag reader / writer; the seller provides the electronic tag tray and the electronic tag meal plate to the diner, and the countertop of the sales window is provided with a sales area electronic tag reader / writer for reading the electronic tag tray and the electronic tag meal plate obtained by the diner. The queue area camera is set at the sales window to clearly collect the position of the image of the diners in the queue, the address of the mobile device obtained by the position collector array, the image information of the diner and the diner information collected by the camera in the door area are compared and matched, and combined with the position at this time The address of the mobile device obtained by the collector array is matched successfully to obtain the corresponding image information, location information, and plate information of the diner in the sales area; when the seller delivers the dishes to the diner, the seller places the dishes within the sensing range of the electronic tag reader in the sales area for reading, and obtains the type and number of dishes of the diner. Since the sales system sells in units of portions, the amount of food taken by the diner for the corresponding dishes is calculated; the location information of the diner from entering the restaurant to the end of picking up the food is extracted and connected to obtain the diner's pre-meal movement line; the number of dishes taken is extracted to obtain the consumption popularity of the dish.

4. The multi-dimensional collection and extraction device for group dining restaurant behavior data according to claim 1 is characterized in that: The catering waste information collection module includes: a leftover collection table, a tray partition board, a leftover collection area camera, a weighing sensor, and a leftover collection electronic tag reader-writer; the tray partition board is composed of vertical boards distributed at continuous intervals, and the distance between two adjacent vertical boards in the tray partition board is greater than the side length of the tray. The tray partition board is fixed on the leftover collection table, and a weighing sensor and a leftover collection electronic tag reader-writer are set between every two vertical boards of the tray partition board and fixed on the leftover collection table. A leftover collection area camera is set above the leftover collection table to collect images of diners who come to place leftover food after dining and the corresponding remaining images of dishes; after dining, diners come to the leftover collection table to place their leftover electronic tag trays and electronic tag plates. The tray partition board is used to constrain the electronic tag tray to be placed above the weighing sensor and ensure that the leftover electronic tag reader-writer can easily read the electronic tag tray information. The leftover collection area camera collects images of diners and images of remaining dishes and matches them with the aforementioned diner preliminary association module and tableware and meal information binding module to form a closed loop of dining information.

5. The multi-dimensional collection and extraction device for group dining restaurant behavior data according to claim 1 is characterized in that: The working process of the first type of tableware and meal information binding module is as follows: after the diner gets the electronic tag tray and a certain number of electronic tag plates, the tableware electronic tag reader obtains the diner's electronic tag tray and electronic tag plate information, and then collects the multiple plate information and tray information sensed at the same time. The tray area camera installed in the buffet area collects the diner's image information and compares and matches it with the diner information collected by the door area camera. Combined with the address of the mobile device obtained by the mobile phone location collector array at this time, after a successful match, the one-to-one corresponding diner image information, location information, and plate information are obtained; during the diner's meal collection process, the tray information read by the dish electronic tag reader is matched with the dish reduction information obtained by the weighing buffet furnace to obtain the diner's meal quantity of the corresponding dish; the diner's location information from entering the restaurant to the end of picking up the meal is extracted and connected to obtain the diner's pre-meal movement line; the reduced number of each dish is extracted to obtain the consumption popularity of the corresponding dish.

6. The multi-dimensional collection and extraction device for group dining restaurant behavior data according to claim 1 is characterized in that: The second tableware and meal information binding module is used for obtaining and binding information of the meal pickup mode as a sales mode.

7. A multi-dimensional collection and extraction method for group dining restaurant behavior data, characterized in that: The data processing process includes: (1) All data are divided into the following dimensions: user characteristics dimension; dish attribute dimension; leftover data dimension; external environment dimension; restaurant environment dimension; transaction data dimension; (2) Transform the dimensioned information into a mathematical model to define a multidimensional data space; assume that the multidimensional data space consists of multiple dimension sets; each dimension has a hierarchical structure, forming a fact table to represent the combined relationship between dimensions and measurements, and each record corresponds to a fact; (3) Data cleaning: processing missing values, detecting outliers, and aligning time series; (4) Slice, dice, drill, rotate, and aggregate the formed data cube; (5) Through mathematical models, the following dimensions of multi-dimensional information are extracted and analyzed: 1) Attribution analysis of food waste, establishing a waste association map including sensory factors such as excessive saltiness, unpleasant taste, greasiness, early satiety, difficulty chewing, and increased dining time, and waste intensity and relative waste index; 2) Attribution analysis of food popularity, through the collected information on diners’ meal pick-up and the consumption of each dish, constructing a prediction and management system for the change of food popularity in different time periods; 3) Restaurant traffic optimization, through the collected diners’ location information and the change process of location, optimizing node layout and path weight, achieving the shortest average moving path of diners, the separation degree between waiters and diners, and minimizing the flow bottleneck of key nodes; 4) Healthy diet assessment, based on the food supply data, standardizing the nutritional content of each dish, marking the cooking method and ingredient composition; 5) Waiter scheduling optimization, based on the collected waiter information and the change process of location, achieving the shortest average moving path of diners, the separation degree between waiters and diners, and minimizing the flow bottleneck of key nodes; Location information and its changing patterns, comparing the number information, location information, waiting information, and dining time information of diners to define core variables, and taking business hours, service demand, and decision-making on whether waiter S is on duty in time period t; 6) Dining behavior prediction, using time series plus external variables as variables to predict total customer flow; establishing a regression model of sales volume, total number of customers, and time characteristics for each dish to predict dish demand; constructing a user-dish matrix, and predicting unobserved consumption preferences through matrix decomposition to predict consumption preferences; 7) Service capability analysis, using the collected number of diners in time period t, the number of employees in the jth category, the average dining time of diners, the consumption of dishes in the kth category, the discarded amount of dishes in the kth category, and the working hours of employees in the jth category as core input variables, and using the service capability score in time period t, the work efficiency of employees in the jth category, and the resource waste index as output variables to achieve: maximizing customer service efficiency, minimizing resource waste, and balancing workload.

Images