Intelligent facial profile personalized recommendation method and system

Abstract

The application provides a wisdom face file personalized recommendation method and system, first, multi-dimensional face file demand data (including historical consumption, real-time demand and preference change data) and face file supply data (including real-time supply, feature change and consumption feedback data) of a user are collected; a demand evolution track is constructed based on the user demand data, and a demand change trend and a correlation relationship are presented; a supply feature time sequence correlation set is constructed based on the supply data, and a supply feature correlation mode and a change rule are presented; a dynamic demand supply matching rule is generated by calling an artificial intelligence model interactive analysis; the rule is iteratively optimized by using user feedback data, and finally a face file personalized recommendation result is generated, precise dynamic matching of demand and supply is realized, and user satisfaction and face file operation benefit are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital online services, in particular to a smart noodle shop personalized recommendation method and system. BACKGROUND

[0002] In the field of catering services, especially in noodle shop operations, how to provide users with accurate and personalized recommendations that meet their needs is a key issue to improve user satisfaction and noodle shop operational efficiency. Traditional noodle shop recommendation methods have obvious shortcomings.

[0003] On the one hand, traditional recommendations mainly rely on simple user historical consumption records, which can only obtain user preferences from limited past data, and cannot fully consider the multidimensionality and dynamics of user needs. Users' dietary needs will change due to factors such as season, health status, personal mood, etc., and real-time dietary needs are also difficult to accurately capture and record. For example, a user may need high-carbohydrate noodles on a day due to high physical activity, but traditional methods cannot timely perceive the temporary needs.

[0004] On the other hand, for noodle shop supply conditions, traditional methods lack in-depth analysis and dynamic tracking of noodle supply characteristics. Noodle supply in noodle shops will change over time, food supply, chef status, etc., and its characteristics such as taste, texture, and nutritional composition will also fluctuate. Traditional recommendations do not consider the change rules of these supply characteristics and their association with user needs, resulting in recommended noodles that may not meet the current expectations of users or the actual supply capacity of noodle shops, and cannot achieve effective matching of demand and supply. SUMMARY

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a smart noodle shop personalized recommendation method, which comprises:

[0006] Collecting user multi-dimensional noodle shop demand data and noodle shop supply data, the user multi-dimensional noodle shop demand data including user historical noodle shop consumption data, real-time dietary demand data, and dietary preference change data, and the noodle shop supply data including noodle shop real-time noodle supply data, noodle characteristic change data, and noodle consumption feedback data;

[0007] Constructing a user noodle shop demand evolution trajectory based on user multi-dimensional noodle shop demand data, the user noodle shop demand evolution trajectory being used to present the change trend and correlation of user noodle shop demand at different time periods;

[0008] Constructing a noodle shop supply characteristic time sequence correlation set based on noodle shop supply data, the noodle shop supply characteristic time sequence correlation set being used to present the correlation mode and change rule of noodle supply characteristics of the noodle shop at different time periods;

[0009] call an artificial intelligence model to interactively analyze the user face file demand evolution track and the face file supply feature time sequence association set, and generate a dynamic demand supply matching rule;

[0010] Iteratively optimize the dynamic demand supply matching rule by using user feedback data on the output result of the dynamic demand supply matching rule, and generate a face file personalized recommendation result based on the optimized dynamic demand supply matching rule.

[0011] In another aspect, the embodiment of the present application also provides a smart face file personalized recommendation system, which comprises a processor and a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine readable storage medium to realize the above-mentioned method.

[0012] Based on the above aspects, the embodiment of the present application can present the change trend and internal correlation of user demand in different time periods by collecting multi-dimensional face file demand data and face file supply data of users, constructing user face file demand evolution track based on multi-dimensional demand data of users, and deeply understanding the dynamic characteristics of user demand. At the same time, the face file supply feature time sequence association set is constructed based on the face file supply data, the association mode and change rule of the face file supply feature in different time periods are accurately presented, the demand evolution track and the supply feature time sequence association set are interactively analyzed by calling an artificial intelligence model, a dynamic demand supply matching rule is generated, and intelligent and dynamic matching of demand and supply is realized. The matching rule is iteratively optimized by using user feedback data, the recommendation result is continuously adapted to the actual demand of users and the supply of face files, and finally the generated face file personalized recommendation result has high precision and adaptability, and the satisfaction of users to face file services is effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 is the execution flow diagram of the smart face file personalized recommendation method provided by the embodiment of the present application.

[0014] Figure 2 is the schematic diagram of exemplary hardware and software components of the smart face file personalized recommendation system provided by the embodiment of the present application. DETAILED DESCRIPTION

[0015] The present application will be specifically described below in conjunction with the drawings of the specification, Figure 1 is the flow diagram of the smart face file personalized recommendation method provided by an embodiment of the present application, and the smart face file personalized recommendation method will be described in detail below.

[0016] Step S110: Collecting user multi-dimensional face profile demand data and face profile supply data, the user multi-dimensional face profile demand data including user historical face profile consumption data, real-time dietary demand data and dietary preference change data, and the face profile supply data including real-time face product supply data, face product feature change data and face product consumption feedback data.

[0017] The embodiment takes the "AI noodle restaurant" chain intelligent noodle restaurant as the unified application scene, covers online applet ordering, offline store dining, delivery and other service modes, and the face products cover three categories of soup noodles (clear soup, thick soup), mixed noodles (sesame paste, onion oil) and fried noodles (sauce, spicy), and the supporting snacks include marinated meat and small dishes. Data is collected through multi-module cooperation, as follows.

[0018] The user multi-dimensional face profile demand data is obtained through a three-layer collection mechanism: the historical face profile consumption data is synchronized from the member system, the cash register system and the delivery platform interface by the consumption record module, and each record includes the consumption period (accurate to the hour), the consumption channel identifier, the face product name, the face product feature combination (taste, texture, ingredients, portion), the consumption amount, the payment method, the evaluation content (text + star level), the repurchase identifier and the like; the real-time dietary demand data is collected by the real-time interaction module through the applet input box, the voice recognition component and the scene perception sensor, for example, the user inputs "light soup noodles, no coriander", the voice recognition "a non-spicy mixed noodle", and the scene perception infers "quick breakfast demand" through positioning (office area), time (7:30-9:00) and device status (low mobile device power); the dietary preference change data is generated by the preference analysis module through historical consumption comparison, user active setting (applet "preference management" page) and evaluation keyword extraction, such as comparing the user from frequently ordering spicy noodles to ordering clear soup noodles, the user checking the "low salt" option, and the evaluation containing the expression "recently like light taste" and the like.

[0019] The face profile supply data is collected through the supply chain and the operation system: the real-time face product supply data is provided by the kitchen management module, including the list of face products on sale in each store, the remaining supply, the estimated serving time, the ingredient inventory status (such as "coriander out of stock" and "sufficient beef inventory"); the face product feature change data is recorded by the product management module, covering seasonal adjustment (adding cold noodles in summer and warm soup noodles in winter), food material replacement (local green replaced by out-of-town green) and process upgrade (hand-pulled noodles instead of machine-pulled noodles), and each record contains the change time, the comparison of the before and after features, and the change reason; the face product consumption feedback data is integrated by the feedback processing module, and the sources include applet evaluation, store comment card and delivery platform comment, and each feedback contains the timestamp, the face product ID, the feedback category (taste / portion / service), the specific content and the associated consumption record ID.

[0020] In terms of privacy protection, user mobile phone numbers and ID numbers (for member registration) are desensitized, with the middle segment of the mobile phone number replaced by asterisks and the ID number retaining only the birthday and gender code; payment information only stores transaction serial numbers without retaining complete card numbers; data transmission uses SSL encryption, sensitive fields are encrypted using the AES algorithm, and the key is managed by a hardware security module; access permissions are divided by role, only data processing and recommendation modules can access desensitized data, and operation logs are recorded throughout the process.

[0021] When collecting user multi-dimensional noodle profile demand data, the consumption record module synchronizes data from the member system, cash register system, and takeout platform interface. Before this, the user has given explicit authorization when registering as a member and placing an order. Through a clear and understandable user agreement, the purpose, scope, and usage of data collection are explained to the user. After the user checks the agreement, data collection begins. When the real-time interaction module collects data through the applet, voice recognition component, and scene perception sensor, it also informs the user of data collection matters in advance in the applet privacy policy, and the user's continued use of the applet is considered consent. For the data acquisition of the preference analysis module, a pop-up window reminds the user that the data will be used for preference analysis and service optimization when the user actively sets preferences or submits evaluations. After the user confirms, the relevant data is recorded.

[0022] The collection of noodle profile supply data is also compliant. Before collecting data, the back-of-house management module, product management module, and feedback processing module ensure that all data sources are legal and that user privacy involved in noodle characteristics changes and consumer feedback has been desensitized through internal regulations and employee training. In terms of data authorization, AI noodle restaurant and third-party data cooperation platforms sign compliant data sharing agreements, clearly defining data usage boundaries and security responsibilities to prevent data misuse. At the same time, professional legal teams are invited to review data collection and authorization processes regularly, updating them to match the dynamic adjustments of laws and regulations to ensure that data operates within legal and compliant frameworks throughout its lifecycle.

[0023] Step S120: Based on the user multi-dimensional noodle profile demand data, a user noodle profile demand evolution trajectory is constructed, which is used to present the change trend and correlation of the user's noodle profile demand in different time periods.

[0024] Taking the demand data of AI noodle restaurant user A as a sample, an evolution trajectory is constructed through data extraction, time axis construction, trend analysis, and other processes. The specific steps are as follows.

[0025] Step S121: From the user multi-dimensional noodle profile demand data, the consumption time data, consumption noodle type data, consumption noodle characteristic data, and consumption evaluation data in the user historical noodle profile consumption data are extracted.

[0026] Four types of core data are extracted from the historical consumption database of user A: consumption time data for the past half year, including specific consumption time points recorded by year, month, day, and hour; consumption noodle type data, which classifies each consumption type into three categories: soup noodles, mixed noodles, and fried noodles; consumption noodle characteristic data, which records the taste (clear soup, spicy, and onion oil), texture (tender, soft and rotten), ingredients (beef, chicken, and vegetables), and portion size (regular, large) of each ordered noodle; and consumption evaluation data, which includes star ratings and text evaluations after each consumption. The text evaluation needs to extract keywords such as "fresh," "spicy," and "adequate portion."

[0027] The extracted data is sorted by consumption time to generate a structured historical consumption list, with each record associated with a unique consumption ID for easy tracing.

[0028] Step S122: Extract current dining scene data, current dining demand description data, and current dining preference tendency data from the real-time dietary demand data of the user's multi-dimensional noodle demand data.

[0029] Extract user A's real-time demand data: Current dining scene data is generated by the scene perception module and includes scene type (office breakfast, community lunch, and family dinner) and scene attributes (time urgency: high, medium, and low; and number of companions: one and multiple), such as "scene type: office breakfast, time urgency: high, number of companions: one." Current dining demand description data is the original text input by the user through the applet, such as "want a light soup noodle, no coriander, and make it quickly." Current dining preference tendency data is generated by extracting keywords from real-time demand description and recent consumption records, such as "light," "soup noodle," "no coriander," and "quick."

[0030] Integrate the three types of data into a real-time demand structure, including data generation timestamp, scene identification, demand text, and preference keyword list.

[0031] Step S123: Extract preference change time node data, preference change before and after comparison data, and preference change influencing factor data from the dietary preference change data of the user's multi-dimensional noodle demand data.

[0032] Extract the preference change data of user A by the preference analysis module: the preference change time node data is the consumption time of the first appearance of preference transition, such as "the time of replacing spicy noodles with clear soup noodles for the first time"; the comparison data before and after the preference change is classified according to the feature dimension, such as the taste dimension (before the change: mainly spicy; after the change: mainly clear soup), the ingredient dimension (before the change: no rejected ingredients; after the change: multiple notes "no coriander"); the preference change influencing factor data is extracted through evaluation text, user active feedback, and scene change, such as mentioning "want to eat light food when the weather is hot" in the evaluation, adding "coriander allergy" in the preference setting, and the season changing from spring to summer, etc.

[0033] Establish an association file for each change node, including consumption samples before and after the change, influencing factor sources, and credibility labels.

[0034] Step S124: Arrange the consumption time data in chronological order, establish a time axis framework, and map the consumption noodle type data, consumption noodle feature data, and consumption evaluation data corresponding to the time point to the corresponding position of the time axis framework.

[0035] Build the demand time axis framework of user A with "week" as the time unit: the horizontal axis is the continuous time node (from the first week to the last week in the past half year), and the vertical axis is divided into three sub-dimensions of type distribution, feature distribution, and evaluation performance. Classify the consumption time data by week, and each time node corresponds to all consumption records of the week; the consumption noodle type data is mapped to the consumption proportion of each type of noodles (such as 80% of soup noodles and 20% of mixed noodles in a week); the consumption noodle feature data is classified and distributed according to taste, texture, ingredients, and portion, respectively (such as 90% of clear soup and 10% of spicy in the taste distribution in a week); the consumption evaluation data is mapped to the average star rating and high-frequency evaluation keywords of the week (such as an average star rating of 4.8 and keywords "light" "fresh" "no coriander").

[0036] The time axis framework uses an extensible data structure for storage, supporting data query and filtering by time node and feature dimension.

[0037] Step S125: Label the current dietary scene data, current dietary demand description data, and current dietary preference tendency data in the real-time dietary demand data at the latest time node position of the time axis framework.

[0038] Label the real-time demand data at the latest time node of the time axis (the last day of the last week in the past half year): the current dietary scene data is displayed in the form of labels, such as "office breakfast" "high urgency"; the current dietary demand description data is associated in the form of a floating text box, and the complete content can be viewed by clicking; the current dietary preference tendency data is presented in the form of a keyword cloud, and the size of the keywords reflects the weight, such as "light" "soup noodles" "no coriander" with the highest weight, displayed in the largest font size.

[0039] The marked data is distinguished from historical data, facilitating intuitive identification of the correlation between real-time demand and historical trends.

[0040] Step S126: Analyze the preference change time node data in the dietary preference change data, mark the preference change time node at the corresponding position of the time axis framework, and associate the preference change before and after the preference change time node.

[0041] Locate the preference change time node in the time axis and mark it with a vertical red line; mark the preference data before the change on the left side of the red line, including type proportion, feature distribution, and evaluation keywords (such as high frequency of “spicy” and “mixed noodles” keywords before the change); mark the preference data after the change on the right side of the red line (such as high frequency of “clear soup” and “noodle soup” keywords after the change); connect the changes in corresponding feature dimensions through arrows, such as “taste: spicy — clear soup” and “ingredients: with parsley — no parsley”.

[0042] Each change node can be clicked to expand a detailed comparison table containing information such as consumption frequency and evaluation star level changes before and after the change.

[0043] Step S127: Based on the data of each time node in the time axis framework, calculate the change rate of the consumption noodle type, the similarity of the consumption noodle feature, and the fluctuation value of the consumption evaluation between adjacent time nodes.

[0044] For two adjacent time nodes in the time axis (such as the Nth week and the N+1th week), calculate the quantitative indicators according to the following logic:

[0045] Step S1271: Extract the consumption noodle type data of the two adjacent time nodes in the time axis framework, and count the number and specific types of the consumption noodle types in the previous time node, and the number and specific types of the consumption noodle types in the next time node.

[0046] Extract the consumption noodle types of the previous node (the Nth week), record the number and specific types (such as 2 types: noodle soup, mixed noodles); extract the consumption noodle types of the next node (the N+1th week), record the number and specific types (such as 2 types: noodle soup, fried noodles).

[0047] Step S1272: Calculate the number of newly added consumption noodle types and the number of disappeared consumption noodle types in the next time node compared with the previous time node, and divide the sum of the number of newly added consumption noodle types and the number of disappeared consumption noodle types by the number of consumption noodle types in the previous time node to obtain the change rate of the consumption noodle type.

[0048] The number of newly added types (e.g., 1 type: fried noodles) and the number of disappeared types (e.g., 1 type: mixed noodles) are counted, and the sum of the two is calculated. The sum is divided by the number of types in the previous node to obtain the type change rate. The higher the change rate, the more significant the change in types.

[0049] Step S1273: Extract the consumption noodle feature data of the adjacent two time nodes, which includes noodle taste features, noodle texture features, noodle ingredient features, and noodle quantity features.

[0050] Extract the feature data of the previous node and the next node in the four dimensions of taste, texture, ingredients, and quantity. For example, the previous node is mainly spicy, and the next node is mainly clear soup; the previous node contains coriander, and the next node does not contain coriander, etc.

[0051] Step S1274: For each noodle feature, convert the noodle feature data of the adjacent time nodes into standardized feature vectors, calculate the similarity value between the two standardized feature vectors, and take the average of the similarity values corresponding to all noodle features as the similarity of the consumption noodle features.

[0052] Convert the data of each feature dimension into a standardized vector (vector elements are the proportion of each feature value), such as a taste vector [spicy proportion, clear soup proportion, onion oil proportion...]; calculate the cosine similarity of the feature vectors corresponding to the previous and next nodes to obtain the similarity of each dimension; take the average of the four dimension similarities as the overall feature similarity. The lower the similarity, the more obvious the feature change.

[0053] Step S1275: Extract the consumption evaluation data of the adjacent two time nodes, which includes evaluation scores, the number of positive words in the evaluation text, and the number of negative words in the evaluation text.

[0054] Extract the average evaluation score, the number of positive words (such as "fresh", "good", "satisfied", etc.), and the number of negative words (such as "salty", "slow", "little", etc.) of the previous node and the next node.

[0055] Step S1276: Calculate the difference between the evaluation score of the next time node and the evaluation score of the previous time node, and calculate the difference between the ratio of the number of positive words to the number of negative words of the next time node and the ratio of the number of positive words to the number of negative words of the previous time node.

[0056] Calculate the difference in evaluation scores (subtract the score of the previous node from the score of the next node); calculate the ratio of the number of positive words to the number of negative words of the previous and next nodes, and then calculate the difference between the two ratios (if the number of negative words is 0, set the ratio according to the preset rule).

[0057] Step S1277: Weighted sum the evaluation score difference and the vocabulary ratio difference, and the weighted weight is set according to the importance of the evaluation score and the vocabulary ratio in the consumption evaluation, to obtain the fluctuation value of the consumption evaluation.

[0058] According to the business experience, the weights of the evaluation score and the vocabulary ratio are set (for example, the score weight is 0.6 and the ratio weight is 0.4), the two difference values are multiplied by the weights respectively and then summed, to obtain the evaluation fluctuation value, and the positive fluctuation value indicates that the evaluation is improved, and the negative fluctuation value indicates that the evaluation is decreased.

[0059] Step S1278: Record the calculated consumption face type change rate, consumption face feature similarity and consumption evaluation fluctuation value, and store them in association with the corresponding adjacent time node pair.

[0060] The three calculated indicators are associated with the corresponding time node pair (for example, the Nth week-the N+1th week), and stored in the quantitative indicator database, and each record contains the time node pair, the type change rate, the feature similarity and the evaluation fluctuation value.

[0061] Step S1279: The calculation results of all adjacent time node pairs are summarized to form a set of user face profile demand change quantitative indicators in each time period in the time axis framework.

[0062] All adjacent node pairs in the time axis are traversed, the quantitative indicators are summarized, and the index set of each time period is arranged according to the time period (for example, monthly, quarterly), so as to facilitate the overall analysis of the demand change law.

[0063] Step S128: According to the calculated change rate, similarity and fluctuation value, a user face profile demand change trend curve is constructed, and the horizontal axis of the user face profile demand change trend curve is the time node, and the vertical axis is the demand change quantitative indicator.

[0064] Taking the time node as the horizontal axis, three trend curves are drawn respectively: the type change rate curve (the vertical axis is the change rate), the feature similarity curve (the vertical axis is the similarity), and the evaluation fluctuation value curve (the vertical axis is the fluctuation value). The curve is smoothed to eliminate the fluctuations caused by accidental consumption and highlight the overall change trend. For example, if the type change rate curve appears a peak value at a certain time point, it indicates that the face type selection change is significant in this period; if the feature similarity curve appears a valley value, it indicates that the preference feature change is obvious in this period.

[0065] Step S129: Associate the diet preference change influencing factor data with the fluctuation nodes in the user face profile demand change trend curve, and label the influencing factors corresponding to each fluctuation node.

[0066] Identify the fluctuation nodes (such as type change rate peaks, feature similarity valley, evaluation fluctuation value extreme time nodes) in the trend curve, associate the extracted preference change influencing factors with the fluctuation nodes, and mark them in the form of annotations at the corresponding positions of the curve, such as "fluctuation node: 10th week, influencing factor: weather turns hot, user setting no spicy preference".

[0067] Each influence factor annotation needs to include the source (evaluation extraction / active setting / scene change) and credibility score (based on data source reliability).

[0068] Step S1210: Integrate the time axis framework, user profile demand change trend curve and influence factor annotation to form a user profile demand evolution trajectory, which includes time dimension demand data distribution, user profile demand change trend and change influencing factor correlation information.

[0069] Integrate the time axis framework, three trend curves, and fluctuation node influence factor annotations into a structured evolution trajectory document, which is divided into three parts: time dimension data distribution (presenting type, feature, and evaluation details by week), change trend analysis (interpreting curve trends, such as "after the 10th week, the taste changes from spicy to clear soup, and the evaluation improves synchronously"), and influence factor correlation table (listing each fluctuation node and its corresponding influence factor, source, and credibility).

[0070] The evolution trajectory supports dynamic updating. When new consumption or demand data is generated by the user, the time axis, trend curve, and influence factor correlation information are automatically updated.

[0071] Step S130: Construct a face profile supply feature time sequence correlation set based on face profile supply data, which is used to present the correlation mode and change law of face product supply features in different time periods.

[0072] Taking the supply data of "AI Noodle Shop" in a certain store as a sample, the set is constructed through time interval division, feature extraction, and correlation analysis processes. The specific steps are as follows.

[0073] Step S131: Extract supply time data, supply product name data, supply product ingredient data, supply product making method data, and supply product quantity data from real-time face product supply data in face profile supply data.

[0074] Extract real-time supply data from the store back-of-house management system: supply time data is the start and end time of each operating period on the same day (such as early market 6:00-10:00, lunch 11:00-14:30, evening market 17:00-21:00); supply noodle name data is the complete name of each period on sale noodles (such as "bone soup beef noodles" "green onion oil chicken noodles" "sauce fried noodles"); supply noodle ingredient data is the detailed ingredient list of each noodle (such as "bone soup beef noodles: beef bone soup, hand-pulled noodles, beef slices, green vegetables, ginger slices, green onion sections"); supply noodle making method data records the making process of each noodle (such as "hand-pulled noodles, bone soup slow stew, point-by-point production"); supply noodle quantity data is the remaining supply quantity and total stock quantity of each noodle on the same day.

[0075] Step S132: Extract feature change time data, pre-change noodle feature data, post-change noodle feature data, and feature change reason data from the noodle feature change data in the noodle stall supply data.

[0076] Extract noodle feature change data from the product management module: feature change time data is the specific time of noodle feature adjustment (accurate to day); pre-change noodle feature data is the noodle feature before adjustment (such as "old pickle noodles: traditional pickle, ordinary soup base"); post-change noodle feature data is the noodle feature after adjustment (such as "old pickle noodles: upgraded pickle, bone soup soup base"); feature change reason data explains the adjustment motivation (such as "food upgrade" "user feedback optimization" "seasonal adjustment").

[0077] Step S133: Extract feedback time data, feedback noodle identification data, feedback content data, and feedback corresponding consumption quantity data from the noodle consumption feedback data in the noodle stall supply data.

[0078] Extract noodle consumption feedback data from the feedback processing module: feedback time data is the timestamp of user feedback; feedback noodle identification data is the unique ID of the feedback corresponding noodle; feedback content data is the specific text of user feedback (such as "bone soup beef noodles soup is very fresh" "green onion oil noodles are too salty"); feedback corresponding consumption quantity data is the purchase quantity in the consumption record associated with the feedback.

[0079] Step S134: Divide the supply time data into multiple continuous time intervals in chronological order, and divide according to the business period (morning, noon, evening) and date (each day as an independent time unit) to form a "day-period" two-level time interval structure, such as "August 1st morning", "August 1st noon", "August 2nd morning", etc. For each time interval, statistics of the distribution of supply noodle name data (such as "August 1st morning supply beef noodle soup, onion oil noodles, green vegetable egg noodles"), supply noodle ingredient data (ingredient list arranged according to noodle), supply noodle making method data (record making process according to noodle), and supply noodle quantity data (statistics of total stock, remaining supply, and total sales of each type of noodle) are formed to form a supply data snapshot of each time interval.

[0080] Step S135: For each time interval, analyze the correspondence between the supply noodle ingredient data and the supply noodle making method data to determine the core feature combination of the noodle supply in this time interval.

[0081] For each time interval, the supply data snapshot is analyzed for feature correlation: taking "August 1st morning" as an example, the correspondence between the ingredients of "beef noodle soup" (beef bone soup, hand-pulled noodles, etc.) and the making method (slow-cooked bone soup, hand-pulled noodles) is analyzed to determine that "bone soup bottom + hand-pulled noodles" is the feature combination of this noodle; the correspondence between the ingredients of "onion oil noodles" (onion oil, hand-pulled noodles, cucumber slices) and the making method (hand-pulled noodles, onion oil freshly prepared) is analyzed to determine that "onion oil seasoning + hand-pulled noodles" is the feature combination. The feature combinations of all noodles in this time interval are summarized, and the feature combination with the highest frequency and the most covered noodles is selected as the core feature combination, for example, the core feature combination of "August 1st morning" is "hand-pulled noodles + traditional soup bottom (bone soup, clear soup)", which covers 80% of the supply noodles in this period.

[0082] Step S136: Correspond the feature change time data in the noodle feature change data to the divided time interval, and associate the pre-change noodle feature data, post-change noodle feature data, and feature change reason data in this time interval.

[0083] According to the feature change time data (accurate to the day) in the noodle feature change data, it is matched to the corresponding "day-period" time interval, for example, a noodle feature change occurs at 9 am on August 1st (morning period), then the change is associated with the "August 1st morning" time interval. In the supply data of this time interval, the pre-change noodle feature data (such as "before change: ordinary soup bottom, machine-pulled noodles"), post-change noodle feature data (such as "after change: bone soup bottom, hand-pulled noodles"), and feature change reason data (such as "reason: user feedback to improve taste") are marked to ensure that each feature change in each time interval can be traced back.

[0084] Step S137: Correspond the feedback time data to the divided time intervals, and count the feedback content data type and the feedback quantity proportion corresponding to different feedback surface quality identification data in each time interval.

[0085] According to the timestamp of the feedback time data, each feedback is associated with the corresponding "day-time period" time interval, for example, a feedback is submitted at 12 o'clock on August 1 (lunch time period), which is classified into the "August 1 lunch" time interval. For each time interval, classify according to the feedback surface quality identification data, and count the feedback content data type (such as taste class, quantity class, service class) of each surface quality, for example, the feedback type of "bone soup beef noodles" in "August 1 lunch" is taste class (70%) and quantity class (30%); calculate the proportion of each feedback type in the total feedback quantity of the surface quality, and the proportion of the feedback quantity of each surface quality in the total feedback quantity of the time interval, to form the feedback distribution statistics.

[0086] Step S138: Calculate the similarity of the core feature combination of the surface quality supply in the adjacent time intervals, and determine the inheritance relationship and change range of the surface quality supply features.

[0087] The core feature combination of the adjacent time intervals is analyzed according to the following process:

[0088] Step S1381: Extract the core feature combination of the surface quality supply of the adjacent two time intervals, and each core feature combination of the surface quality supply includes a supply surface quality ingredient feature subset, a supply surface quality making method feature subset, and a supply surface quality quantity distribution feature subset.

[0089] Extract the core feature combination of "August 1 morning" and "August 1 lunch": the former includes an ingredient feature subset (bone soup, clear soup, handmade ramen ingredients), a making method feature subset (handmade ramen, bone soup slow cooking), and a quantity distribution feature subset (bone soup beef noodles accounting for 40%, onion oil mixed noodles accounting for 30%); the latter includes an ingredient feature subset (bone soup, sauce, handmade ramen ingredients), a making method feature subset (handmade ramen, sauce stir-frying), and a quantity distribution feature subset (bone soup beef noodles accounting for 20%, sauce stir-fried noodles accounting for 40%).

[0090] Step S1382: Convert each feature subset in the core feature combination of the surface quality supply of the previous time interval into a feature matrix, and the rows of the feature matrix represent the feature types and the columns represent the specific attribute values of the features.

[0091] Convert the ingredient feature subset of "8 / 1 morning market" into a matrix: rows represent "soup type" "noodle type" "side dish type", and columns represent specific attribute values (e.g. the soup type column contains the proportion of bone soup and clear soup); the cooking method feature subset matrix: rows represent "ramen method" "soup making", columns represent specific attribute values (e.g. the ramen method column contains the proportion of handmade ramen); the quantity distribution feature subset matrix: rows represent "noodle type", columns represent specific proportion values.

[0092] Step S1383: Convert each feature subset in the core feature combination of the noodle supply in the next time interval into a feature matrix consistent with the dimension of the feature matrix of the previous time interval.

[0093] Convert the feature subsets of "8 / 1 noon market" according to the matrix dimension of the previous interval: the soup type column in the ingredient feature matrix contains the proportion of bone soup and sauce, ensuring that the rows are consistent with the previous matrix; add a "stir-frying method" row to the cooking method matrix, but keep the "ramen method" row to ensure dimension matching; the quantity distribution matrix is arranged according to the same noodle type row, and the column value of the corresponding noodle is set to 0.

[0094] Step S1384: For each feature matrix corresponding to a feature subset, use a matrix similarity algorithm to calculate the similarity value of the adjacent time interval feature matrix, which calculates the overall similarity based on the correspondence of matrix elements.

[0095] For the ingredient feature matrix, calculate the element similarity of the corresponding row (such as the difference in bone soup proportion in the soup type), and then take the average value as the matrix similarity; for the cooking method matrix, calculate the element similarity of the corresponding row; for the quantity distribution matrix, calculate the cosine similarity of the noodle proportion.

[0096] Step S1385: Take the average of all feature subset similarity values as the overall similarity of the core feature combination of the noodle supply in the adjacent time interval.

[0097] Take the average of the similarity values of the three feature subsets of ingredients, cooking methods, and quantity distribution to get the overall similarity of the core feature combination of the adjacent interval, for example, the overall similarity between "8 / 1 morning market" and "8 / 1 noon market" is 0.6.

[0098] Step S1386: If the overall similarity is higher than the set inheritance threshold, it is determined that the core feature combination of the noodle supply in the next time interval inherits the core feature combination of the noodle supply in the previous time interval, and the inherited feature subsets and specific features are recorded.

[0099] Set the inheritance threshold to 0.5. Since the overall similarity is 0.6, which is higher than the threshold, it is determined that the afternoon inherits the core feature combination of the morning. The inherited features are recorded: "hand-pulled noodle ingredients" in ingredients, and "hand-pulled noodles" in making method.

[0100] Step S1387: If the overall similarity is lower than the set inheritance threshold, analyze the difference feature subset of the core feature combination of the food supply in the two time intervals, and count the number of feature changes and the type of attribute changes in the difference feature subset.

[0101] If the overall similarity of a certain adjacent interval is 0.4 (lower than the threshold), analyze the difference feature subset: such as the soup type in the ingredient feature subset changes from "bone soup, clear soup" to "spicy soup, tomato soup", and the making method changes from "hand-pulled noodles" to "machine-pulled noodles", and the number of changes is 2 feature subsets, and the attribute type of changes is "soup type" and "making method type".

[0102] Step S1388: Calculate the attribute value change amplitude of the changed features in the difference feature subset. Divide the difference between the attribute value of the next time interval and the attribute value of the previous time interval by the attribute value of the previous time interval to obtain the change amplitude of each changed feature.

[0103] For the change of "soup type", the proportion of bone soup in the previous interval is 40%, and the proportion of spicy soup in the next interval is 50%, the change amplitude = (50%-40%) / 40%; For the change of "making method", the proportion of hand-pulled noodles in the previous interval is 100%, and the proportion of machine-pulled noodles in the next interval is 100%, the change amplitude = (100%-100%) / 100%=0.

[0104] Step S1389: Take the average of the change amplitudes of all changed features as the overall change amplitude of the core feature combination of the food supply in the adjacent time intervals.

[0105] Take the average of the change amplitudes of each individual changed feature to obtain the overall change amplitude, which is used to quantify the difference degree of the supply features in adjacent intervals.

[0106] Step S139: Calculate the association degree between the feedback content data type and the core feature combination of the food supply in each time interval, and determine the user feedback tendency corresponding to different supply features.

[0107] For each time interval, analyze the association of feedback content data types and core feature combinations: for example, the core feature combination of "early market on August 1st" is "hand-pulled noodles + bone soup base", and the proportion of "soup is fresh" and "noodles have chewiness" in taste-related feedback (accounting for 80% of taste-related feedback) and the proportion of "enough" in quantity-related feedback (accounting for 70% of quantity-related feedback) in this interval are counted. The degree of association of these positive feedbacks with the core feature combination (such as the degree of association = the number of positive feedbacks / the total number of feedbacks covered by the core feature combination) is calculated; at the same time, the degree of association of negative feedbacks (such as "soup is salty") with the core feature combination is calculated to determine that the feature combination of "hand-pulled noodles + bone soup base" corresponds to a user feedback tendency of "mainly positive feedback, mainly recognizing taste and soup base flavor".

[0108] Step S1310: Integrate the core feature combinations of the noodle shop supply in each time interval, the feature change information, the feedback association information, and the feature association relationship of adjacent intervals to form a noodle shop supply feature time sequence association set. The noodle shop supply feature time sequence association set includes time interval division, supply features in each interval, feature time sequence association, and feedback association information.

[0109] The "day-time period" time interval division, the core feature combinations of each interval, the feature change record (including the features before and after the change and the reason), the feedback association analysis results (feedback type distribution, degree of association with core features, feedback tendency), the similarity and inheritance / change information of adjacent intervals are integrated to form a structured noodle shop supply feature time sequence association set. The noodle shop supply feature time sequence association set is stored in a time sequence data structure, which supports querying supply features by time interval and querying feedback tendency by feature combination, and directly presents the evolution law of supply features over time and the association mode with user feedback.

[0110] Step S140: Call the artificial intelligence model to interactively analyze the user noodle shop demand evolution trajectory and the noodle shop supply feature time sequence association set, and generate dynamic demand supply matching rules.

[0111] Call the artificial intelligence model (trained based on massive demand supply data) built based on the Transformer architecture to generate matching rules through feature interaction, filtering, rule generation, etc. The specific steps are as follows.

[0112] Step S141: Convert the time axis framework, user noodle shop demand change trend curve, and influencing factor labels in the user noodle shop demand evolution trajectory into demand feature vectors recognizable by the artificial intelligence model.

[0113] Feature encoding is performed on the user demand feature evolution track: the type proportion in the time axis framework is converted into a numerical feature component; the trend feature of the demand change trend curve (such as rising, falling, and stable) is converted into a trend feature component; and the influence factor label is converted into a classification feature component (such as “weather factor” and “preference setting”). The above components are arranged in a preset order to form a fixed-dimension demand feature vector, and each element in the vector corresponds to a demand feature attribute.

[0114] Step S142: converting the time interval division, supply feature in each interval, feature time sequence association, and feedback association information in the supply feature time sequence association set into a supply feature vector recognizable by the artificial intelligence model.

[0115] Feature encoding is performed on the supply feature time sequence association set: the supply core feature combination of the time interval is converted into a feature combination component; the similarity and change amplitude in the feature time sequence association are converted into a time sequence association component; and the association degree and feedback tendency in the feedback association information are converted into a feedback feature component. The above components are arranged in the dimension order corresponding to the demand feature vector to form a supply feature vector, ensuring that the dimension of the supply feature vector matches that of the demand feature vector, facilitating interactive analysis.

[0116] Step S143: inputting the demand feature vector recognizable by the artificial intelligence model and the supply feature vector recognizable by the artificial intelligence model into a feature interaction layer of the artificial intelligence model, and using a cross-attention mechanism in the feature interaction layer to capture the association points of the demand feature and the supply feature in the time dimension.

[0117] The demand feature vector and the supply feature vector are input into the feature interaction layer, and the cross-attention mechanism in the feature interaction layer locates the association points of the demand feature and the supply feature in the time dimension by calculating the attention weights of the demand feature and the supply feature: for example, the time feature of “summer preference for light soup noodles” in the demand feature forms a high-weight association with the time feature of “summer supply of cold noodles and light soup noodles” in the supply feature; and the preference feature of “rejection of coriander” in the demand feature forms an association with the ingredient feature of “lack of coriander” in the supply feature.

[0118] Step S144: calculating the association strength between each element in the demand feature vector recognizable by the artificial intelligence model and each element in the supply feature vector recognizable by the artificial intelligence model through the feature interaction layer to generate an association strength matrix, and the element value in the association strength matrix represents the association degree of the corresponding demand feature and supply feature.

[0119] The association strength matrix is generated according to the following process:

[0120] Step S1441: Determine the number of elements of the demand feature vector recognizable by the artificial intelligence model and the demand feature type corresponding to each element in the demand feature vector recognizable by the artificial intelligence model, the demand feature type including time demand feature, preference demand feature, scene demand feature and evaluation demand feature.

[0121] The number of elements of the demand feature vector is counted (such as 100 elements), and the demand feature type is labeled for each element: elements 1-20 correspond to time demand feature (consumption period distribution), elements 21-50 correspond to preference demand feature (taste, ingredient preference), elements 51-70 correspond to scene demand feature (scene type, time urgency), and elements 71-100 correspond to evaluation demand feature (evaluation star level, keyword distribution).

[0122] Step S1442: Determine the number of elements of the supply feature vector recognizable by the artificial intelligence model and the supply feature type corresponding to each element in the supply feature vector recognizable by the artificial intelligence model, the supply feature type including time supply feature, surface feature, quantity supply feature and feedback supply feature.

[0123] The number of elements of the supply feature vector is counted (consistent with the demand feature vector, 100 elements), and the supply feature type is labeled: elements 1-20 correspond to time supply feature (supply period, feature change time), elements 21-50 correspond to surface feature (ingredients, preparation method), elements 51-70 correspond to quantity supply feature (supply amount, sales distribution), and elements 71-100 correspond to feedback supply feature (feedback type, correlation degree).

[0124] Step S1443: Initialize the correlation strength matrix, the number of rows of the correlation strength matrix is equal to the number of elements of the demand feature vector recognizable by the artificial intelligence model, the number of columns of the correlation strength matrix is equal to the number of elements of the supply feature vector recognizable by the artificial intelligence model, and the initial element values of the correlation strength matrix are all set to an initial correlation reference value.

[0125] A 100x100 correlation strength matrix is constructed, all element initial values are set to 0 (initial correlation reference value), representing that the demand and supply features are not associated in the initial state.

[0126] Step S1444: For each element in the demand feature vector recognizable by the artificial intelligence model, extract the feature attribute, feature numerical range and importance weight of the element in the user face profile demand evolution track.

[0127] For element 25 in the demand feature vector (preference demand feature, corresponding to "light taste preference"), the feature attribute is "taste preference type", the feature value range is "0-1" (0 represents no preference, 1 represents strong preference), and the importance weight is 0.8 (set according to the degree of influence of this feature on demand).

[0128] Step S1445: For each element in the supply feature vector identifiable by the artificial intelligence model, the feature attribute, feature value range, and importance weight of the element in the supply feature time sequence association set are extracted.

[0129] For element 25 in the supply feature vector (surface feature, corresponding to "soup base proportion"), the feature attribute is "soup base type proportion", the feature value range is "0-100%", and the importance weight is 0.7 (set according to the representativeness of this feature to supply).

[0130] Step S1446: Calculate the attribute matching degree and value range overlap degree between the elements in the demand feature vector identifiable by the artificial intelligence model and the elements in the supply feature vector identifiable by the artificial intelligence model, and weight the attribute matching degree, value range overlap degree, importance weight of the elements in the demand feature vector identifiable by the artificial intelligence model, and importance weight of the elements in the supply feature vector identifiable by the artificial intelligence model to obtain the association strength value between a single element in the demand feature vector identifiable by the artificial intelligence model and a single element in the supply feature vector identifiable by the artificial intelligence model.

[0131] Calculate the attribute matching degree (both related to taste / soup base, matching degree is 1) of element 25 (demand) and element 25 (supply); the value range overlap degree (demand element value is 0.9, supply element value is 0.8, overlap degree is 0.8); association strength value = (attribute matching degree x 0.5 + value range overlap degree x 0.5) x (demand weight x 0.5 + supply weight x 0.5), the specific association strength value is obtained through this weighted calculation.

[0132] Step S1447: Fill the calculated association strength value into the corresponding position in the association strength matrix, replacing the initial association reference value in the association strength matrix.

[0133] Fill the above association strength value into the 25th row and 25th column of the association strength matrix, replacing the initial value 0.

[0134] Step S1448: Traverse all elements of the demand feature vector identifiable by the artificial intelligence model and all elements of the supply feature vector identifiable by the artificial intelligence model, repeat the above process of calculating the association strength value, until all elements of the association strength matrix are updated to the calculated association strength value, forming a complete association strength matrix.

[0135] The correlation strength values of all element pairs in the demand feature vector and the supply feature vector are sequentially calculated, the corresponding positions in the matrix are updated, and finally a complete correlation strength matrix is formed, wherein each element value in the matrix represents the correlation degree of a pair of demand-supply features.

[0136] Step S145: input the correlation strength matrix into the feature screening layer of the artificial intelligence model, and use the feature screening layer to retain feature pairs with correlation strength meeting the set condition and eliminate feature pairs with correlation strength not meeting the set condition.

[0137] A correlation strength threshold (such as 0.5) is set, and the feature screening layer traverses all elements in the correlation strength matrix: retain feature pairs with element values ≥ 0.5 (such as “light taste preference-clear soup soup bottom proportion” and “no coriander preference-coriander out-of-stock state”), eliminate feature pairs with element values < 0.5 (such as “spicy taste preference-clear soup soup bottom proportion” and “fast demand-slow braising cooking method”), and record the correlation strength values and corresponding demand-supply feature types of the retained feature pairs to form a screened feature pair list. The screened feature pair list only contains feature combinations with strong correlation.

[0138] Step S146: classify the retained feature pairs, divide the feature pairs into different matching categories according to the demand feature type and the supply feature type, and each matching category corresponds to a matching relationship between a type of demand and supply.

[0139] According to the combination of the demand feature type (time demand, preference demand, scene demand, evaluation demand) and the supply feature type (time supply, surface feature, quantity supply, feedback supply) in the retained feature pairs, classify: classify the feature pairs of “light taste preference-clear soup soup bottom proportion” and “no coriander preference-coriander stock state” and other “preference demand-surface feature” combinations into the “preference-feature matching category”; classify the feature pairs of “morning market demand-morning market supply surface list” and “fast demand-short service time surface” and other “time / scene demand-time / quantity supply” combinations into the “scene-supply matching category”; and classify the feature pairs of “high evaluation demand-high feedback score surface” and other “evaluation demand-feedback supply” combinations into the “evaluation-feedback matching category”. Each matching category contains multiple feature pairs, and the feature pairs in the same category have similar demand-supply matching logic.

[0140] Step S147: input the classified feature pairs into the rule generation layer of the artificial intelligence model, and use the rule generation layer to generate initial matching rules containing feature correlation conditions, matching priorities, and matching result output forms based on the attributes of the feature pairs in each matching category.

[0141] The rule generation layer generates an initial rule for each matching category: for the "preference-feature matching category", generate a rule such as "if the user has a light taste preference (demand feature) and the face product supply has a clear soup base ratio ≥ set proportion (supply feature), then preferentially match this type of face product" based on the attribute of the feature, wherein the feature association condition is "light taste preference identification = yes and clear soup base ratio ≥ set proportion", the matching priority is set to the highest (because preference directly affects user selection willingness), and the matching result output form is "face product name + clear soup feature annotation"; for the "scene-supply matching category", generate a rule such as "if the user is in the early market fast demand scene (demand feature) and the face product estimated serving time ≤ set time (supply feature), then match this type of face product", the feature association condition is "scene type = early market and time urgency = high and serving time ≤ set time", the matching priority is set to medium, and the output form is "face product name + serving time annotation"; for the "evaluation-feedback matching category", generate a rule such as "if the user prefers high evaluation face products (demand feature) and the face product historical feedback score ≥ set level (supply feature), then match this type of face product", the feature association condition is "high evaluation preference = yes and feedback score ≥ set level", the matching priority is set to the second highest, and the output form is "face product name + score annotation".

[0142] Step S148: Apply the initial matching rule to the historical demand-supply data sample to obtain the matching result of the initial matching rule on the historical demand-supply data sample, and analyze the fit degree of the matching result of the initial matching rule on the historical demand-supply data sample and the historical actual consumption result, and adjust the feature association condition and the matching priority in the initial matching rule.

[0143] Select a historical demand-supply data sample (containing demand data and corresponding face product supply data, actual consumption records of multiple users) in the past half year, apply the initial matching rule to the sample: for the "light taste + early market fast demand" data of user B in the sample, apply the rule to match "clear soup vegetable noodles (fast serving, clear soup base)", and the actual consumption result is "clear soup vegetable noodles", which has a high fit degree; for the "light taste + high evaluation demand" data of user C, apply the rule to match "clear soup beef noodles (high score)", but the actual consumption is "light chicken noodles", which has a low fit degree. Analyze the reason for the low fit degree: the initial rule does not consider the implicit demand of user C "preference for chicken ingredients", and the feature association condition is not perfect. Accordingly, adjust the rule: supplement the "ingredient preference" association condition in the "preference-feature matching category" rule, and optimize the "light taste preference-clear soup base ratio" rule to "light taste preference = yes and clear soup base ratio ≥ set proportion and ingredients contain user preferred ingredients"; at the same time, based on the sample fit degree data, adjust the "preference-feature matching category" priority from "highest" to "higher than the scene category", to ensure that it is more in line with the actual consumption habit.

[0144] Step S149: Repeat the process of adjusting the initial matching rule until the fit degree of the initial matching rule on the historical demand supply data sample reaches a stable state, and determine the initial matching rule reaching the stable state as the dynamic demand supply matching rule.

[0145] Repeat the "apply rule-analyze fit degree-adjust rule" process: Apply the optimized rule to the historical sample again. If the fit degree improves and the fit degree fluctuation range is ≤ the set threshold after continuous application for multiple times (indicating that the rule tends to be stable), stop adjusting. The finally determined dynamic demand supply matching rule includes three types of optimized sub-rules: "preference-feature matching sub-rule" (including multi-dimensional preference association conditions such as taste, ingredients, and texture, with the highest priority), "scene-supply matching sub-rule" (including time period, urgency, and other scene association conditions, with medium priority), and "evaluation-feedback matching sub-rule" (including score, feedback keywords, and other association conditions, with the second highest priority), and each sub-rule can be dynamically adapted according to the demand supply data changes.

[0146] Step S150: Iteratively optimize the dynamic demand supply matching rule using the feedback data of the user on the output result of the dynamic demand supply matching rule, and generate a personalized recommendation result for the noodles based on the optimized dynamic demand supply matching rule.

[0147] Optimize the rule continuously through user feedback and generate the final recommendation result, the specific steps are as follows.

[0148] Step S151: Obtain the initial recommendation result output by the dynamic demand supply matching rule, which includes the recommended noodle name, recommended noodle feature description, and recommended basis.

[0149] Input the demand data of the current user and the noodle supply data into the dynamic demand supply matching rule, and output the initial recommendation result: For example, for the "light, no coriander, early market fast demand" of user A, the recommendation result includes "recommended noodle 1: vegetable noodle with light soup (features: handmade noodles, light soup, no coriander, short delivery time; recommended basis: meets the light taste preference, no coriander requirement, and early market fast supply condition)" and "recommended noodle 2: chicken noodle with light soup (features: handmade noodles, chicken soup, no coriander, high score; recommended basis: meets the preference and high evaluation demand)".

[0150] Step S152: Present the initial recommendation result to the user and collect the feedback operation data of the user on the initial recommendation result, which includes the user's selection data on the recommended noodles, the user's evaluation data on the recommended basis, and the user's supplementary demand description data.

[0151] The initial recommendation result is presented in the form of a list through the applet, and user feedback is collected: the user selects "vegetable noodles with clear soup" (selection data), evaluates the recommendation as "accurate description of delivery time, but no mention of portion size" (evaluation data), and supplements the demand description as "need regular portion size" (supplementary data); if the user does not select the recommended noodles and inputs "want noodles with some chewiness", the selection data is "no selection of recommended noodles", and the supplementary demand description is "prefer chewy noodles".

[0152] Step S153: Classify and organize the feedback operation data, mark the recommended noodles selected by the user as positive feedback noodles, and mark the recommended noodles not selected by the user as non-positive feedback noodles, extract the evaluation keywords of the recommended basis and the keywords of the user's supplementary demand description.

[0153] Classify and organize the feedback data: mark "vegetable noodles with clear soup" as positive feedback noodles, and "chicken noodle soup" as non-positive feedback noodles; extract the keywords "missing portion size" from the evaluation data and the keywords "regular portion size" and "chewy noodles" from the supplementary description. At the same time, record the time of feedback occurrence, user identification and corresponding recommendation result ID for correlation analysis.

[0154] Step S154: Analyze the feature association conditions in the dynamic demand supply matching rules corresponding to the positive feedback noodles, and count the frequency and association strength of the feature association conditions in the positive feedback cases.

[0155] Analyze the rule feature association conditions corresponding to the positive feedback noodles "vegetable noodles with clear soup": "light taste = yes" "no coriander = yes" "delivery time ≤ set value". Count the frequency of the above three conditions in all positive feedback cases in the past week: "light taste = yes" has the highest frequency, "delivery time ≤ set value" follows; calculate the association strength of each condition with positive feedback (association strength = number of positive cases containing the condition / total number of positive cases), and determine "light taste" and "no coriander" as strong association conditions.

[0156] Step S155: Analyze the feature association conditions in the dynamic demand supply matching rules corresponding to the non-positive feedback noodles, find the contradictions between the feature association conditions and user feedback, and determine the feature association conditions in the dynamic demand supply matching rules that need to be adjusted.

[0157] Analyze the rule conditions corresponding to the non-positive feedback dish "chicken noodle soup noodles": "light taste = yes" "no cilantro = yes" "score ≥ set level". Combine the user's unselected feedback and supplementary demand "tender noodles" to find the contradiction: the type of noodles in this dish is "ordinary ramen", which does not meet the user's implicit "tender" texture demand, and the initial rule does not contain the "texture preference" association condition. Accordingly, it is determined that the rule needs to be adjusted: add a "texture preference" feature association condition in the "preference-feature matching sub-rule".

[0158] Step S156: Convert the user's supplementary demand description keywords into new demand features, determine the association relationship between the new demand features and the existing supply features, and supplement the association relationship to the dynamic demand supply matching rule.

[0159] Convert the supplementary demand keywords "regular portion" and "tender noodles" into new demand features: "portion preference = regular" and "texture preference = tender". Determine the association relationship between the new demand features and the supply features: "portion preference = regular" is associated with "dish portion = regular" supply feature, and "texture preference = tender" is associated with "noodle type = handmade ramen (tender type)" supply feature. Supplement the above association relationship to the "preference-feature matching sub-rule", and the optimized rule condition is "light taste = yes and no cilantro = yes and portion = regular and noodle type = handmade ramen (tender type)".

[0160] Step S157: Adjust the association strength of the feature association conditions in the dynamic demand supply matching rule, increase the association strength of the feature association conditions that appear more frequently than the set frequency in positive feedback cases, and reduce the association strength of the feature association conditions that exist in contradiction with non-positive feedback.

[0161] According to the positive feedback case statistics, the association strength of "light taste" and "no cilantro" conditions is increased from the original set value (because the frequency is higher than the set frequency); for the condition "score ≥ set level" in non-positive feedback, there is no direct association with user selection, the association strength of this condition is reduced to avoid over-reliance on scores and ignore key preferences such as texture.

[0162] Step S158: Adjust the matching priority in the dynamic demand supply matching rule, and increase the matching category priority of the positive feedback dish, and reduce the matching category priority of the non-positive feedback dish.

[0163] Based on the user's selection of "vegetable noodle soup" (corresponding to "preference-feature matching class") and non-selection of "chicken noodle soup" (corresponding to "evaluation-feedback matching class"), the priority of "preference-feature matching class" is further increased to "highest", higher than "evaluation-feedback matching class", to ensure that the user's core preferences are met first.

[0164] Step S159: Apply the adjusted dynamic demand supply matching rule to the new demand supply data sample to obtain a matching result of the adjusted dynamic demand supply matching rule on the new demand supply data sample, and collect feedback data of the user on the matching result.

[0165] Select a new user demand supply data sample (e.g., the "light, chewy, regular portion demand" of user D), apply the adjusted rule, and match out "handmade chewy clear soup noodles (regular portion)". Present the result to the user and collect feedback: the user selects the noodles and evaluates "meets demand", which is positive feedback.

[0166] Step S1510: Compare the user feedback satisfaction of the adjusted dynamic demand supply matching rule and the adjusted dynamic demand supply matching rule on the same data sample. If the user feedback satisfaction corresponding to the adjusted dynamic demand supply matching rule is higher than the user feedback satisfaction corresponding to the adjusted dynamic demand supply matching rule, keep the adjusted dynamic demand supply matching rule. If the user feedback satisfaction corresponding to the adjusted dynamic demand supply matching rule is lower than the user feedback satisfaction corresponding to the adjusted dynamic demand supply matching rule, adjust the parameters of the dynamic demand supply matching rule in the reverse direction until the user feedback satisfaction corresponding to the dynamic demand supply matching rule reaches a set level.

[0167] Compare the satisfaction and optimize according to the following process:

[0168] Step S1510-1: Select the demand supply data sample applied by the adjusted dynamic demand supply matching rule as the comparison sample set, which includes the demand data of multiple users and the corresponding noodle supply data.

[0169] Select 100 user demand supply samples applied by the rule before adjustment as the comparison sample set, covering different demand types (preferences, scenarios, evaluation demands) and supply scenarios.

[0170] Step S1510-2: Apply the adjusted dynamic demand supply matching rule to the comparison sample set to generate the adjusted recommendation result of the adjusted dynamic demand supply matching rule on each sample in the comparison sample set.

[0171] Apply the adjusted rule to each of the 100 samples to generate an adjusted recommendation result for each sample, such as recommending "handmade chewy clear soup noodles" for a "light, no coriander, chewy demand" user in the sample, and recommending "ordinary clear soup noodles" before adjustment.

[0172] Step S1510-3: Retrieve the pre-adjustment recommendation results generated by the adjusted dynamic demand supply matching rule on the comparison sample set and the user feedback data collected at the time.

[0173] The recommended result and feedback data of the pre-adjustment rule applied to the sample set are retrieved from the database, such as the user selection rate of "ordinary clear soup noodles" is 60%, and the satisfaction evaluation is "general".

[0174] Step S1510-4: Collecting feedback data of the user on the adjusted recommended result, which includes recommended noodle selection rate, recommended basis recognition degree, and demand satisfaction score.

[0175] The feedback of the adjusted recommended result is collected: the selection rate of "handmade chewy clear soup noodles" is 85%, the recommended basis recognition degree (the proportion of users who think that the recommended reason meets the demand) is 90%, and the demand satisfaction score (1-5 points) is 4.5 on average.

[0176] Step S1510-5: Calculating the user feedback satisfaction degree corresponding to the pre-adjustment dynamic demand supply matching rule, which is based on the pre-adjustment recommended noodle selection rate, pre-adjustment recommended basis recognition degree, and pre-adjustment demand satisfaction score, using weighted summation method, and the weighted weight is set according to the importance of each index.

[0177] Set the selection rate weight to 0.4, the recognition degree weight to 0.3, and the satisfaction score weight to 0.3, and calculate the pre-adjustment satisfaction degree = 60% x 0.4 + 70% x 0.3 + 3.5 x 0.2 (converted to 0-1 interval after scoring 1-5) = pre-adjustment satisfaction value.

[0178] Step S1510-6: Calculate the user feedback satisfaction degree corresponding to the adjusted dynamic demand supply matching rule, which uses the same weighted summation method as the user feedback satisfaction degree corresponding to the pre-adjustment dynamic demand supply matching rule, based on the adjusted recommended noodle selection rate, the adjusted recommended basis recognition degree, and the adjusted demand satisfaction score.

[0179] Calculate the post-adjustment satisfaction degree = 85% x 0.4 + 90% x 0.3 + 4.5 x 0.2 = post-adjustment satisfaction value with the same weight.

[0180] Step S1510-7: Compare the user feedback satisfaction degree of the pre-adjustment dynamic demand supply matching rule and the user feedback satisfaction degree of the post-adjustment dynamic demand supply matching rule corresponding to each sample in the comparison sample set, and count the number of samples with improved user feedback satisfaction degree, the number of samples with decreased user feedback satisfaction degree, and the number of samples with unchanged user feedback satisfaction degree.

[0181] Statistical results: 80 samples have improved satisfaction, 15 samples remain unchanged, and 5 samples have decreased.

[0182] Step S1510-8: Calculate the overall user feedback satisfaction improvement amplitude, which is the sum of the difference between the user feedback satisfaction of the adjusted dynamic demand-supply matching rule and the user feedback satisfaction of the unadjusted dynamic demand-supply matching rule of all samples in the comparison sample set divided by the total number of samples in the comparison sample set.

[0183] Calculate the improvement amplitude = (sum of all sample satisfaction difference) / 100 = improvement amplitude value.

[0184] Step S1510-9: If the overall user feedback satisfaction improvement amplitude is greater than the set improvement threshold, and the proportion of the number of samples whose user feedback satisfaction has improved in the total number of samples in the comparison sample set is greater than the set proportion threshold, then it is determined that the adjusted dynamic demand-supply matching rule is superior to the unadjusted dynamic demand-supply matching rule.

[0185] If the improvement amplitude value is greater than the set threshold (e.g., 5%), and the proportion of improved samples (80%) is greater than the set proportion threshold (70%), then it is determined that the adjusted rule is superior and is retained.

[0186] Step S1510-10: If the above conditions are not met, record the demand-supply characteristics of the samples whose user feedback satisfaction has decreased, and analyze the reasons why the adjusted dynamic demand-supply matching rule is not applicable to these samples.

[0187] If the conditions are not met, analyze the characteristics of the 5 decreased samples: for example, all of them are "high score priority" demand users, and the adjusted rule reduces the score condition weight, resulting in matching deviation, so the condition weight needs to be adjusted in the opposite direction.

[0188] Step S1511: Generate a personalized recommendation result based on the optimized dynamic demand-supply matching rule, and the specific process is as follows.

[0189] For example, step S1511-1: Convert the multi-dimensional face profile demand data of the current user into a current demand feature vector that meets the input format of the optimized dynamic demand-supply matching rule.

[0190] Convert the real-time demand of user A ("light, no coriander, chewy, early market fast, regular portion") into a demand feature vector, including "light flavor = 1, no coriander = 1, chewy = 1, early market = 1, time urgency = 1, portion = 1", etc. The format is consistent with the rule input requirements.

[0191] Step S1511-2: Convert the supply data of the current face profile into a current supply feature vector that meets the input format of the optimized dynamic demand-supply matching rule.

[0192] Convert the current supply data of the store ("Hand-made thick soup noodles: clear soup base, no coriander, regular portion, short delivery, high score; chicken noodle soup: clear soup base, no coriander, ordinary ramen, medium delivery, high score") into a supply feature vector, including features such as taste, ingredients, texture, delivery time, score, etc.

[0193] Step S1511-3: input the current demand feature vector and the current supply feature vector into the optimized dynamic demand-supply matching rule, which filters out the supply features matching the current demand features according to the feature association conditions.

[0194] The rule filters out the supply features of "hand-made thick soup noodles" according to the condition "light = 1 and no coriander = 1 and texture = 1 and portion = 1 and delivery = 1".

[0195] Step S1511-4: according to the matching priority in the optimized dynamic demand-supply matching rule, sort the noodles corresponding to the filtered supply features, and extract the detailed information of the top K noodles, which includes the name of the noodles, the ingredients of the noodles, the making method of the noodles, the taste description of the noodles and the suitable dining scene of the noodles.

[0196] According to the optimized matching priority (preference-feature matching class > scene-supply matching class > evaluation-feedback matching class), the filtered "hand-made thick soup noodles", "thick vegetable soup noodles" and "quick handmade soup noodles" are sorted: "hand-made thick soup noodles" ranks first because it meets all the preferences and scene conditions of "light, no coriander, thick, regular portion, short delivery"; "thick vegetable soup noodles" ranks second because it lacks the "short delivery" scene condition; "quick handmade soup noodles" ranks third because the ingredients do not contain vegetables, which is the user's potential preference. Extract the detailed information of the top 3 (K = 3) noodles: "hand-made thick soup noodles" - ingredients: handmade noodles, beef bone soup, green vegetables, ginger; making method: handmade noodles, slow cooking with bone soup and quick packaging; taste description: light and fresh, no irritating seasoning; suitable scene: office breakfast, single quick meal. "Thick vegetable soup noodles" - ingredients: handmade noodles, chicken bone soup, various vegetables, green onions; making method: handmade noodles, clear soup cooking; taste description: light and refreshing, vegetables with strong aroma; suitable scene: community lunch, healthy light food demand. "Quick handmade soup noodles" - ingredients: handmade noodles, concentrated soup, eggs, green vegetables; making method: handmade noodles, ready-to-eat; taste description: light and convenient, thick texture; suitable scene: takeout delivery, time-constrained meal.

[0197] Step S1511-5: According to the matching basis in the optimized dynamic demand supply matching rule, generate a recommendation reason for each recommended noodle dish, which contains specific matching points between the noodle dish and the user's demand characteristics.

[0198] Generate a recommendation reason for the first-ranked "handmade chewy clear soup noodles": "The noodle dish uses handmade noodles, meeting your preference for 'chewy texture'; the soup base is beef bone clear soup, meeting the 'light' taste requirement; the ingredients do not contain cilantro, matching your 'no cilantro' requirement; the serving size is regular, adapting to your serving size requirement; the serving time is short, fitting the'morning market fast' scenario, with the highest overall matching degree." Generate a recommendation reason for "chewy vegetable clear soup noodles": "The noodle dish is handmade ramen, meeting the 'chewy texture' preference; the chicken bone clear soup meets the 'light' taste, without cilantro; the ingredients contain a variety of seasonal vegetables, providing rich nutrition; but the serving time is slightly longer than the first recommendation, with a lower overall matching degree." Generate a recommendation reason for "fast handmade clear soup noodles": "The noodle dish is handmade ramen, meeting the 'chewy texture' preference; the concentrated clear soup meets the 'light' taste, without cilantro; the serving speed is fast, adapting to the 'fast' scenario; but the variety of vegetables in the ingredients is less, with a third overall matching degree."

[0199] Step S1511-6: Organize the detailed information and recommendation reasons of the recommended noodle dishes into a structured recommendation list in the order of ranking, with each record in the structured recommendation list containing the complete information of a recommended noodle dish and the corresponding recommendation reason.

[0200] The structured recommendation list is organized in the format of "ranking position-noodle dish information-recommendation reason": Position 1 - Noodle dish name: Handmade chewy clear soup noodles; Ingredient composition: handmade noodles, beef bone clear soup, green vegetables, ginger slices; Production method: handmade ramen, slow simmering of bone soup and quick packaging; Taste description: light and fragrant, no irritating seasonings; Suitable scenario: office breakfast, single fast meal; Recommendation reason: [see corresponding content in step S1511-5 for details]. Position 2 - Noodle dish name: Chewy vegetable clear soup noodles; Ingredient composition: handmade noodles, chicken bone clear soup, a variety of seasonal vegetables, green onions; Production method: handmade ramen, clear soup cooking; Taste description: light and refreshing, with rich vegetable aroma; Suitable scenario: community lunch, healthy light food demand; Recommendation reason: [see corresponding content in step S1511-5 for details]. Position 3 - Noodle dish name: Fast handmade clear soup noodles; Ingredient composition: handmade noodles, concentrated clear soup, eggs, green vegetables; Production method: handmade ramen pre-prepared, heated and eaten; Taste description: light and convenient, chewy texture; Suitable scenario: takeout delivery, time-constrained meal; Recommendation reason: [see corresponding content in step S1511-5 for details].

[0201] Step S1511-7: The face product information in the structured recommendation list is integrated, and the integrated structured recommendation list is determined as the noodle shop personalized recommendation result, which contains recommended face products sorted by matching priority, detailed information of each recommended face product, and corresponding recommendation reasons.

[0202] The face product information in the structured recommendation list is integrated according to three dimensions of “basic information-feature matching-scene adaptation”: the basic information dimension summarizes the face product name, ingredients, and making method; the feature matching dimension extracts the matching items of each face product and user preferences (light, no coriander, and chewy); and the scene adaptation dimension explicitly specifies the suitable dining scenes and delivery characteristics. After integration, a complete noodle shop personalized recommendation result is formed, ensuring clear information hierarchy, which contains not only the detailed content of individual face products but also the matching differences of different face products.

[0203] Step S1511-8: The noodle shop personalized recommendation result is output in a displayable form.

[0204] The output form is selected according to the user's use scenario: if the user orders through the applet, a card list is displayed, each recommended face product corresponds to a card, the top of the card displays the ranking and “recommended” mark, the middle shows the face product name, taste, and suitable scene, and the bottom presents the recommendation reasons and “add to cart” button; if the user operates the self-service ordering machine in the store, a split-screen form is displayed, the left side is the recommendation list, and the right side is the detailed information and picture display of the selected face product, supporting touch screen sliding to switch face products; if it is a delivery platform, a simple text list is sent to the user's message center, each recommendation contains the face product name, core matching points, and jump link, and clicking can directly enter the ordering page. During the output process, the face product pictures, ingredient information, etc. in the recommendation result are checked for compliance to ensure that there is no false advertising content, and the user's ordering preferences and other privacy data are protected and not disclosed to third parties.

[0205] Figure 2 A schematic diagram of exemplary hardware and software components of a smart noodle shop personalized recommendation system 100 that can implement the idea of the present application is shown. For example, a processor 120 can be used in the smart noodle shop personalized recommendation system 100 and used to perform the functions in the present application.

[0206] The smart noodle shop personalized recommendation system 100 can be a general-purpose server or a special-purpose server, both of which can be used to implement the smart noodle shop personalized recommendation method of the present application. Although only one server is shown in the present application, for the sake of convenience, the functions described in the present application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0207] For example, the smart face profile personalized recommendation system 100 can include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, a ROM, or a RAM, or any combination thereof. Illustratively, the smart face profile personalized recommendation system 100 can also include program instructions stored in a ROM, a RAM, or other types of non-transitory storage media, or any combination thereof. The methods of the present application can be implemented according to the program instructions. The smart face profile personalized recommendation system 100 also includes an I / O interface 150 between the computer and other input and output devices.

[0208] For ease of illustration, only one processor is described in the smart face profile personalized recommendation system 100. However, it should be noted that the smart face profile personalized recommendation system 100 in the present application can also include multiple processors, and therefore the steps performed by one processor described in the present application can also be jointly performed by multiple processors or individually performed by multiple processors. For example, if the processor of the smart face profile personalized recommendation system 100 performs steps A and B, it should be understood that steps A and B can also be jointly performed by two different processors or individually performed in one processor. For example, a first processor performs step A, a second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0209] In addition, the present application also provides a readable storage medium, wherein computer executable instructions are pre-stored in the readable storage medium, and when the processor executes the computer executable instructions, the smart face profile personalized recommendation method is realized.

[0210] It should be noted that, in order to simplify the description of the present application and to help understand one or more embodiments of the present application, in the foregoing description of the embodiments of the present application, various features are sometimes combined into one embodiment, drawing, or description thereof.

Claims

1. A personalized recommendation method for smart noodle stalls, characterized in that, The method includes: Collect multi-dimensional noodle stall demand data and noodle stall supply data from users. The multi-dimensional noodle stall demand data from users includes historical noodle stall consumption data, real-time dietary demand data, and dietary preference change data. The noodle stall supply data includes real-time noodle product supply data, noodle product characteristic change data, and noodle product consumption feedback data. Based on multi-dimensional user profile demand data, a user profile demand evolution trajectory is constructed. This trajectory is used to present the changing trends and correlations of user profile demand in different time periods. Based on the noodle stall supply data, a time-series correlation set of noodle stall supply characteristics is constructed. This time-series correlation set of noodle stall supply characteristics is used to present the correlation patterns and changing rules of noodle stall supply characteristics in different time periods. The artificial intelligence model is invoked to perform interactive analysis on the evolution trajectory of user faceplate demand and the time-series correlation set of faceplate supply characteristics, and to generate dynamic demand and supply matching rules. The dynamic demand and supply matching rules are iteratively optimized by using user feedback data on the output results of the dynamic demand and supply matching rules, and personalized recommendation results are generated based on the optimized dynamic demand and supply matching rules. The iterative optimization of the dynamic demand-supply matching rule using user feedback data on the output results of the dynamic demand-supply matching rule includes: Obtain the initial recommendation results output by the dynamic demand and supply matching rules. These initial recommendation results include the recommended noodle names, descriptions of recommended noodle features, and the basis for the recommendation. The initial recommendation results are presented to the user, and the user's feedback data on the initial recommendation results is collected. This feedback data includes the user's selection data of the recommended noodle products, the user's evaluation data of the basis for the recommendation, and the user's supplementary description of needs. The feedback data is categorized and organized. Recommended noodles selected by users are marked as positive feedback noodles, and recommended noodles not selected by users are marked as negative feedback noodles. Keywords for users' evaluation of the basis for the recommendations and keywords for users' supplementary descriptions of their needs are extracted. Analyze the feature association conditions in the dynamic demand and supply matching rules corresponding to positive feedback noodle products, and statistically analyze the frequency and association strength of the feature association conditions in positive feedback cases; Analyze the feature association conditions in the dynamic demand and supply matching rules corresponding to non-positive feedback noodle products, find the contradictions between the feature association conditions and user feedback, and determine the feature association conditions in the dynamic demand and supply matching rules that need to be adjusted. The user-supplemented demand description keywords are converted into new demand features, the correlation between the new demand features and existing supply features is determined, and the correlation is added to the dynamic demand and supply matching rules. Adjust the correlation strength of feature correlation conditions in the dynamic demand and supply matching rules, increase the correlation strength of feature correlation conditions that occur more frequently than the set frequency in positive feedback cases, and decrease the correlation strength of feature correlation conditions that contradict non-positive feedback. Adjust the matching priority in the dynamic demand and supply matching rules, increase the priority of the matching category corresponding to positive feedback noodle products, and decrease the priority of the matching category corresponding to non-positive feedback noodle products; The adjusted dynamic demand and supply matching rules are applied to the new demand and supply data samples to obtain the matching results of the adjusted dynamic demand and supply matching rules on the new demand and supply data samples, and user feedback data on the matching results are collected. Compare the user feedback satisfaction with the dynamic demand and supply matching rules before and after the adjustment on the same data sample. If the user feedback satisfaction corresponding to the adjusted dynamic demand and supply matching rules is higher than that corresponding to the original dynamic demand and supply matching rules, then the adjusted dynamic demand and supply matching rules are retained. If the user feedback satisfaction corresponding to the adjusted dynamic demand and supply matching rules is lower than that corresponding to the original dynamic demand and supply matching rules, then the parameters of the dynamic demand and supply matching rules are adjusted in reverse until the user feedback satisfaction corresponding to the dynamic demand and supply matching rules reaches the set level.

2. The personalized recommendation method for smart noodle stalls according to claim 1, characterized in that, The construction of the user profile requirement evolution trajectory based on multi-dimensional user profile requirement data includes: Extract consumption time data, noodle type data, noodle characteristic data, and consumption evaluation data from users' multi-dimensional noodle demand data; Extract current dietary scenario data, current dietary demand description data, and current dietary preference data from real-time dietary demand data from multi-dimensional user demand data; Extract data on the time points of preference changes, comparison data before and after preference changes, and data on factors influencing preference changes from multi-dimensional user demand data; Arrange the consumption time data in chronological order to establish a timeline framework, and map the consumption type data, consumption characteristic data, and consumption evaluation data of the corresponding time points to the corresponding positions in the timeline framework. Mark the current dietary scenario data, current dietary demand description data, and current dietary preference data in the real-time dietary demand data at the latest time node position in the timeline frame; Analyze the time nodes of preference changes in the dietary preference change data, mark the time nodes of preference changes in the corresponding positions of the time axis frame, and associate them with the comparison data of preference changes before and after the time nodes of preference changes; Based on the data at each time point in the time axis framework, the rate of change of consumer noodle types, the similarity of consumer noodle characteristics, and the fluctuation value of consumer evaluation are calculated between adjacent time points. Based on the calculated rate of change, similarity, and fluctuation value, a trend curve of user profile demand change is constructed. The horizontal axis of the trend curve of user profile demand change is the time node, and the vertical axis is the quantitative indicator of demand change. The data on factors influencing changes in dietary preferences are correlated with the fluctuation nodes in the trend curve of changes in user noodle demand, and the influencing factors corresponding to each fluctuation node are labeled. By integrating the timeline framework, the trend curve of user profile demand changes, and the annotation of influencing factors, a user profile demand evolution trajectory is formed. This trajectory includes the distribution of demand data over time, the trend of user profile demand changes, and the correlation information of influencing factors.

3. The personalized recommendation method for smart noodle stalls according to claim 2, characterized in that, The calculation of the rate of change of consumer noodle types, the similarity of consumer noodle characteristics, and the fluctuation value of consumer evaluation between adjacent time nodes, based on data from each time node in the time axis framework, includes: Extract the consumption noodle product type data between two adjacent time nodes in the timeline frame, count the number and specific types of noodle product types consumed at the previous time node, and count the number and specific types of noodle product types consumed at the next time node. Calculate the number of new noodle product types and the number of noodle product types that disappeared at the next time point compared to the previous time point. Divide the sum of the number of new noodle product types and the number of noodle product types that disappeared by the number of noodle product types at the previous time point to obtain the rate of change of noodle product types. Extract the consumption feature data of noodles from two adjacent time points. This consumption feature data includes noodle flavor features, noodle texture features, noodle ingredient features, and noodle portion features. For each type of noodle product feature, the noodle product feature data of adjacent time nodes are converted into standardized feature vectors, the similarity value between two standardized feature vectors is calculated, and the average of the similarity values ​​corresponding to all noodle product features is taken as the similarity of the consumption noodle product features. Extract consumer review data from two adjacent time points. This consumer review data includes review scores, the number of positive words in the review text, and the number of negative words in the review text. Calculate the difference between the evaluation score at the next time point and the evaluation score at the previous time point, and calculate the difference between the ratio of the number of positive words to the number of negative words at the next time point and the ratio of the number of positive words to the number of negative words at the previous time point. The difference between the evaluation scores and the difference between the vocabulary ratios are weighted and summed. The weighting is set according to the importance of the evaluation scores and vocabulary ratios in the consumer evaluation, and the fluctuation value of the consumer evaluation is obtained. Record the calculated rate of change of consumer noodle product type, similarity of consumer noodle product characteristics, and fluctuation value of consumer evaluation, and store them in association with the corresponding adjacent time nodes; The calculation results of all adjacent time node pairs are summarized to form a set of quantitative indicators of changes in user profile requirements for each time period in the time axis framework.

4. The personalized recommendation method for smart noodle stalls according to claim 1, characterized in that, The construction of a time-series correlation set of surface-area supply characteristics based on surface-area supply data includes: Extract the supply time data, noodle name data, noodle ingredient data, noodle preparation method data, and noodle quantity data from the noodle stall supply data in real-time noodle supply data; Extract the time data of feature changes, the feature data of the product before changes, the feature data of the product after changes, and the data on the reasons for feature changes from the noodle supply data. Extract feedback time data, noodle product identification data, feedback content data, and corresponding consumption quantity data from the noodle supply data; The supply time data is divided into multiple consecutive time intervals in chronological order, and the distribution of the supply noodle product name data, supply noodle product ingredient data, supply noodle product preparation method data, and supply noodle product quantity data in each time interval is statistically analyzed. For each time interval, analyze the correspondence between the data on the ingredients of the supplied noodles and the data on the methods of producing the supplied noodles, and determine the core characteristic combination of the noodles supply within that time interval; Map the feature change time data in the noodle product feature change data to the divided time interval, and associate the noodle product feature data before the change, the noodle product feature data after the change, and the feature change reason data within the time interval. The feedback time data is mapped to the divided time intervals, and the data types and proportions of feedback content corresponding to different feedback product identification data are statistically analyzed within each time interval. Calculate the similarity of core feature combinations of noodle supply within adjacent time intervals to determine the inheritance relationship and variation range of noodle supply features; Calculate the correlation between the data type of feedback content and the core feature combination of noodle supply within each time interval, and determine the user feedback tendency corresponding to different supply features; By integrating the core feature combination, feature change information, feedback correlation information, and feature correlation relationship of noodle supply in each time interval, a noodle supply feature time sequence correlation set is formed. This noodle supply feature time sequence correlation set includes time interval division, supply features of each interval, feature time sequence correlation, and feedback correlation information.

5. The personalized recommendation method for smart noodle stalls according to claim 4, characterized in that, The calculation of the similarity of core feature combinations of noodle supply within adjacent time intervals, and the determination of the inheritance relationship and variation range of noodle supply features, includes: Extract the core feature combination of noodle supply between two adjacent time intervals. Each core feature combination of noodle supply includes a subset of noodle ingredient features, a subset of noodle preparation method features, and a subset of noodle quantity distribution features. Each subset of features in the core feature combination of the noodle supply in the previous time interval is converted into a feature matrix. The rows of the feature matrix represent the feature type, and the columns represent the specific attribute values ​​of the features. Transform each feature subset in the core feature combination of the noodle supply in the next time interval into a feature matrix with the same dimension as the feature matrix of the previous time interval. For each feature subset, a matrix similarity algorithm is used to calculate the similarity value of feature matrices in adjacent time intervals. This matrix similarity algorithm calculates the overall similarity based on the correspondence between matrix elements. The average of the similarity values ​​of all feature subsets is taken as the overall similarity of the core feature combination of noodle supply in adjacent time intervals. If the overall similarity is higher than the set inheritance threshold, it is determined that the core feature combination of the noodle supply in the later time interval has inherited the core feature combination of the noodle supply in the previous time interval, and the inherited feature subset and specific features are recorded. If the overall similarity is lower than the set inheritance threshold, then analyze the difference feature subsets of the core feature combination of the noodle supply in the two time intervals, and count the number of feature changes and the type of feature attributes in the difference feature subsets. Calculate the magnitude of attribute value change for each variable feature in the subset of variable features. Divide the difference between the attribute value in the next time interval and the attribute value in the previous time interval for each variable feature by the attribute value in the previous time interval to obtain the magnitude of change for a single variable feature. The average value of the changes in all the changing characteristics is taken as the overall change in the combination of core characteristics of noodle supply in adjacent time intervals. Record the similarity of core feature combinations of noodle supply in adjacent time intervals, the results of inheritance relationship determination, and the overall change range, and associate them with the corresponding time interval of the noodle supply feature time sequence association set.

6. The personalized recommendation method for smart noodle stalls according to claim 1, characterized in that, The process involves using an artificial intelligence model to interactively analyze the evolution trajectory of user faceplate demand and the time-series correlation set of faceplate supply characteristics, generating dynamic demand-supply matching rules, including: The timeline framework, trend curve of user profile demand change, and influencing factor annotation in the user profile demand evolution trajectory are converted into demand feature vectors that can be recognized by artificial intelligence models. The time interval division, supply characteristics of each interval, feature time sequence correlation and feedback correlation information in the face stock supply feature time sequence correlation set are converted into supply feature vectors that can be recognized by artificial intelligence models. The demand feature vector and the supply feature vector that can be identified by the artificial intelligence model are input into the feature interaction layer of the artificial intelligence model. The feature interaction layer uses a cross-attention mechanism to capture the correlation points between demand features and supply features in the time dimension. The correlation strength matrix is ​​generated by calculating the correlation strength between each element in the demand feature vector that the artificial intelligence model can identify and each element in the supply feature vector that the artificial intelligence model can identify through the feature interaction layer. The element values ​​in the correlation strength matrix represent the degree of correlation between the corresponding demand feature and the supply feature. The correlation strength matrix is ​​input into the feature filtering layer of the artificial intelligence model. The feature filtering layer retains feature pairs whose correlation strength meets the set conditions and removes feature pairs whose correlation strength does not meet the set conditions. The retained feature pairs are classified and divided into different matching categories according to the type of demand feature and the type of supply feature. Each matching category corresponds to a type of demand and supply matching relationship. The classified feature pairs are input into the rule generation layer of the artificial intelligence model. The rule generation layer uses the feature pair attributes of each matching category to generate initial matching rules that include feature association conditions, matching priorities, and matching result output formats. The initial matching rule is applied to the historical demand and supply data sample to obtain the matching result of the initial matching rule on the historical demand and supply data sample. The degree of fit between the matching result of the initial matching rule on the historical demand and supply data sample and the historical actual consumption result is analyzed, and the feature association conditions and matching priority in the initial matching rule are adjusted. The process of repeatedly adjusting the initial matching rules continues until the initial matching rules reach a stable state in terms of their fit with historical demand and supply data samples. The initial matching rules that have reached a stable state are then identified as dynamic demand and supply matching rules.

7. The personalized recommendation method for smart noodle stalls according to claim 6, characterized in that, The step of calculating the correlation strength between each element in the demand feature vector recognizable by the AI ​​model and each element in the supply feature vector recognizable by the AI ​​model through the feature interaction layer, and generating a correlation strength matrix, includes: Determine the number of elements in the demand feature vector that the artificial intelligence model can recognize and the demand feature type corresponding to each element in the demand feature vector that the artificial intelligence model can recognize. The demand feature type includes time demand features, preference demand features, scenario demand features and evaluation demand features. Determine the number of elements in the supply feature vector that the artificial intelligence model can recognize and the type of supply feature corresponding to each element in the supply feature vector that the artificial intelligence model can recognize. The type of supply feature includes time supply feature, product feature, quantity supply feature and feedback supply feature. Initialize the association strength matrix. The number of rows in the association strength matrix is ​​equal to the number of elements in the demand feature vector that the artificial intelligence model can identify. The number of columns in the association strength matrix is ​​equal to the number of elements in the supply feature vector that the artificial intelligence model can identify. The initial element values ​​of the association strength matrix are all set to the initial association baseline values. For each element in the demand feature vector that can be identified by the artificial intelligence model, extract the feature attributes, feature value range, and importance weight of the element in the user profile demand evolution trajectory. For each element in the supply feature vector that can be identified by the artificial intelligence model, extract the feature attributes, feature value range, and importance weight of the element in the temporal correlation set of surface supply features. The attribute matching degree and numerical range overlap of the elements in the demand feature vector that can be identified by the artificial intelligence model and the supply feature vector that can be identified by the artificial intelligence model are calculated. The attribute matching degree, numerical range overlap and the importance weight of the elements in the demand feature vector that can be identified by the artificial intelligence model and the importance weight of the elements in the supply feature vector that can be identified by the artificial intelligence model are weighted and calculated to obtain the correlation strength value of a single element in the demand feature vector that can be identified by the artificial intelligence model and a single element in the supply feature vector that can be identified by the artificial intelligence model. Fill the calculated association strength value into the corresponding position in the association strength matrix, replacing the initial association reference value in the association strength matrix; Iterate through all elements of the demand feature vector and the supply feature vector that the AI ​​model can identify, repeating the above process of calculating the correlation strength value until all elements of the correlation strength matrix are updated with the calculated correlation strength value, forming a complete correlation strength matrix.

8. The personalized recommendation method for smart noodle stalls according to claim 1, characterized in that, The comparison of user satisfaction with the dynamic demand-supply matching rules before and after the adjustment on the same data sample includes: The demand and supply data samples that were applied under the dynamic demand and supply matching rules before the adjustment were selected as the comparison sample set. This comparison sample set contains demand data and corresponding surface supply data of multiple users. The adjusted dynamic demand and supply matching rules are applied to the comparison sample set to generate the adjusted recommendation results for each sample in the comparison sample set. Retrieve the pre-adjustment recommendation results generated on the comparison sample set by the dynamic demand and supply matching rules before the adjustment, as well as the user feedback data collected at that time. Collect user feedback data on the adjusted recommendation results. This feedback data includes the selection rate of recommended noodle products, the degree of acceptance of the recommendation basis, and the score of demand satisfaction. Calculate the user feedback satisfaction corresponding to the dynamic demand-supply matching rule before the adjustment. The user feedback satisfaction calculation is based on the recommended noodle selection rate before the adjustment, the recognition of the recommendation basis before the adjustment, and the demand satisfaction score before the adjustment. The weighted summation method is used, and the weight is set according to the importance of each indicator. Calculate the user feedback satisfaction corresponding to the adjusted dynamic demand and supply matching rule. The user feedback satisfaction is calculated using the same weighted summation method as the user feedback satisfaction corresponding to the original dynamic demand and supply matching rule, based on the adjusted recommended noodle selection rate, the adjusted recognition of the recommendation basis, and the adjusted demand satisfaction score. Compare the user feedback satisfaction with the dynamic demand-supply matching rule before and after the adjustment for each sample in the sample set, and count the number of samples with improved user feedback satisfaction, decreased user feedback satisfaction, and unchanged user feedback satisfaction. The overall improvement in user feedback satisfaction is calculated as the sum of the differences between the user feedback satisfaction of the adjusted dynamic demand supply matching rule and the user feedback satisfaction of the unadjusted dynamic demand supply matching rule across all samples in the comparison sample set, divided by the total number of samples in the comparison sample set. If the overall improvement in user feedback satisfaction exceeds the set improvement threshold, and the proportion of the number of samples with improved user feedback satisfaction to the total number of samples in the comparison sample set exceeds the set proportion threshold, then the adjusted dynamic demand supply matching rule is determined to be superior to the original dynamic demand supply matching rule. If the above conditions are not met, record the demand and supply characteristics corresponding to the samples where user feedback satisfaction has decreased, and analyze the reasons for the inapplicability of the adjusted dynamic demand and supply matching rules in the samples.

9. A smart noodle stall personalized recommendation system, characterized in that, The device includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the smart face card personalized recommendation method according to any one of claims 1-8.

Images