Method for checking differences between user self-filled washing requirements and standard process

Abstract

The application discloses a user self-filling washing requirement and standard process difference checking method, and particularly relates to the field of life technology. Based on a process knowledge base, a clothes characteristic vector is searched by a standard process searching algorithm to obtain a corresponding standard process parameter vector, a secondary searching index of the knowledge base is constructed, the process knowledge base is iterated, a candidate set satisfying a constraint is obtained, a record with the highest score is screened out, a recommended process and a safety boundary are obtained. The application is realized based on a standard parameter vector, a rule engine and a statistical learning method, can be conveniently integrated into an existing Internet washing platform or a traditional washing management system, and can be flexibly extended according to a standard process library of different enterprises, and is easy to integrate and expand.

Description

Technical Field

[0001] This invention relates to the field of daily life technology, specifically to a method for verifying the differences between user-defined washing requirements and standard processes. Background Technology

[0002] With the fast pace of life and the widespread adoption of online lifestyle service platforms, the internet-based laundry business is developing rapidly. Users place orders through online service platforms (such as apps, mini-programs, or websites), filling in the type and quantity of clothing and basic requirements. The platform then assigns the order to a pickup rider or the nearest store. The clothes are then transported to a laundry factory, where they undergo sorting, pre-treatment, main washing, rinsing, spin-drying, drying / air-drying, ironing, and quality inspection according to internal processes. Finally, the clothes are returned to the user through a delivery network, and quality feedback is recorded in the system. Throughout this process, factories typically develop a relatively fixed washing process based on clothing labels and internal standard processes to ensure consistent quality and safety for large-volume orders.

[0003] Online lifestyle service platforms offer users a wide range of personalized laundry options. Besides selecting basic service types (such as "dry cleaning," "washing," and "simple ironing"), internet laundry users can also express detailed requirements through text notes or by selecting optional parameters. Examples include: safety preferences: "Absolutely no bleach"; "No high-temperature drying, fear of shrinkage"; "As gentle as possible"; effect preferences: "Heavy oil stains on the collar, want it cleaned thoroughly"; "Mold spots, want them removed as much as possible"; health and experience preferences: "Children / sensitive skin at home, no strong fragrances"; "Dust mite allergy, want high-temperature sterilization"; cost and time considerations: "Can be slower, but must be clean"; "In a hurry, hope it's washed and returned within three days."

[0004] These requirements from clients have the following characteristics:

[0005] 1. Mostly natural language, unstructured: Users are used to expressing themselves in spoken language, such as "Don't use too much force" or "Wash it cleaner," which are vague and difficult to map directly to specific process parameters.

[0006] 2. May conflict with labels and standard processes: For example, if the user requests a 60℃ high-temperature wash, the garment label may specify a maximum water temperature of 30℃; if the user requests a strong spin-drying + high-temperature drying, the garment material may be wool / silk / down; if the user requests an extremely low temperature and extremely gentle wash, the stain type may be stubborn oil stains, and the standard process recommends a higher intensity, which may result in incomplete stain removal.

[0007] 3. Limited user safety awareness: Users often only focus on "cleanliness" and "thorough sterilization and mite removal", but they are unaware of the limits that clothing fibers can withstand. They are prone to making excessive demands on the process, which may lead to the risk of damaging the clothes.

[0008] 4. Existing systems do not adequately support self-filling requirements: Most platforms only store "order remarks" as strings in the database, which are then read and judged manually by store or factory employees. This has the following limitations: it is easy to miss or misunderstand; different employees have different experience and risk tolerance, leading to inconsistent judgments; and it is difficult to convert the data into analyzable data, making it impossible to form a systematic difference assessment and optimization.

[0009] In addition, the existing methods for verifying discrepancies in laundry orders have the following problems:

[0010] 1. Front-end options limitations:

[0011] For example, the water temperature selection dropdown menu does not offer a "temperature higher than the label requirement" option;

[0012] If the user selects "high temperature mite removal", the system will automatically prompt "some materials do not support high temperature".

[0013] This approach is too crude and can only prevent the most obvious conflicts. It cannot finely distinguish different types of risks and their degree of difference, nor can it handle complex situations (such as washing multiple garments together or requiring specific stains on certain parts).

[0014] 2. The back-end relies on manual review:

[0015] Some platforms delegate the review of orders with complex notes to customer service or factory auditors on a case-by-case basis. This review process relies heavily on experience, and different auditors may provide different solutions for the same requirement. Manual review suffers from low efficiency, a high risk of missed checks, inconsistent standards, and difficulty in scaling.

[0016] 3. Lack of a unified model for process variability:

[0017] Existing practices mostly involve "rule-triggered warnings," such as "a pop-up reminder is triggered if the label prohibits drying but the user requests drying." These practices do not establish a unified difference calculation model in the multi-dimensional parameter space, making it impossible to quantify the "degree of deviation" and "comprehensive risk."

[0018] 4. Lack of data closure and adaptive optimization:

[0019] Even though the platform records some complaints and rewash information, it rarely forms a systematic learning mechanism for these results and process differences, resulting in it remaining at the level of patching things up based on experience after many years of operation. Summary of the Invention

[0020] Therefore, this invention uses the enterprise standard washing process knowledge base as a benchmark and the user-filled washing requirements as variables to quantify the differences between the two within a unified parameter space, thereby achieving automatic verification and risk control. It provides a method for verifying the differences between user-filled washing requirements and standard processes to solve the problems mentioned in the background art.

[0021] To achieve the above objectives, the present invention provides the following technical solution: a method for verifying the difference between user-filled washing requirements and standard processes, the method comprising: based on a process knowledge base, processing garment feature vectors... The corresponding standard process parameter vector is obtained through the standard process retrieval algorithm. A secondary retrieval index is constructed in the knowledge base, and then the process knowledge base is traversed to obtain a candidate set that meets the constraints and the record with the highest score is selected to obtain the recommended process and safety boundary.

[0022] The clothing feature vector G includes user-selected options and user-filled process parameters. ; Standard process parameter vectors are analyzed through semantic parsing mathematical models. User-filled process parameters Perform parameter dimension unification and coding standardization; calculate the single-dimensional difference degree one by one through the difference degree calculation model. Combining the weights of each process parameter dimension To obtain the overall difference It combines thresholds to classify risk levels and identify risk types; at the same time, it makes hard conflict judgments through a conflict rule verification model to identify safety risks, quality risks, cost waste and experience deviations, and provides a basis for judgment to generate difference results.

[0023] Based on the discrepancy results, an explanation template and several alternative process solutions are generated to serve the user. The final solution and subsequent complaints and rewashing situations are recorded. Weights and thresholds are adjusted periodically using historical data to form a closed-loop discrepancy verification method driven by knowledge base, rules and data.

[0024] Preferably, define clothing feature vectors. ;

[0025] in, The clothing category is indicated by user selection or store entry.

[0026] Indicates the fabric material;

[0027] Indicates the color type;

[0028] Indicates information about filling and auxiliary materials;

[0029] This represents a set of washing label symbols, which is a binary vector obtained through photo recognition or manual input. It contains a total of [number missing]. One tag;

[0030] Standard process parameter vectors are obtained by predefining the process knowledge base and retrieving them based on clothing feature vectors. ;

[0031] in, The standard washing water temperature is indicated in °C and is derived from the process knowledge base and washing labels. The standard mechanical strength grade is derived from the process knowledge base; This indicates the standard main wash time, in minutes. Indicates the standard dehydration speed rating; Indicates a bleaching permit mark; Indicates the standard drying method code; Indicates whether it includes a high-temperature disinfection / mite removal function; Indicates the method of using fabric softener; This indicates the recommended unit cost or energy consumption level;

[0032] User-inputted process parameters ;

[0033] in, This indicates the user's requested washing water temperature; This indicates the user's required mechanical strength grade; This indicates the user's requested main wash time; This indicates the user's required dehydration speed level; This indicates the user's bleaching request; This indicates the user's requested drying method; This indicates the user's requirement for high-temperature disinfection / mite removal; This indicates the user's preference for fabric softener; This indicates the user's expected unit cost or energy consumption level; making and Dimensional alignment and consistent value selection rules are maintained; if no explicit user requirements are specified, the standard process will be executed by default. ;in, Indicates the parameter dimension;

[0034] The rules for selecting values ​​are as follows:

[0035] First, unify data types:

[0036] Numerical data: washing water temperature T, main wash time S, process cost / energy consumption level C, all stored as floating-point numbers;

[0037] Graded type: Mechanical strength M, dehydration speed grade D, fabric softener usage F, unified integer code;

[0038] Bleaching type B and drying method R are allowed, and a unified integer encoding mapping table is used.

[0039] Boolean type: High-temperature disinfection / mite removal H, unified integer encoding;

[0040] The boundary value handling mechanism is as follows:

[0041] Single-dimensional boundary constraints:

[0042] For numerical / rank parameters:

[0043] ;

[0044] For each parameter dimension Define its safe value range Made of fabric material Set of washing label symbols Decision; among them, For the first User values ​​for each parameter dimension;

[0045] For enumeration / Boolean parameters:

[0046] ;

[0047] in, This is the set of legal values ​​for this dimension; For the first Standard values ​​for each parameter dimension; This represents the first value in the user process parameter vector after boundary handling and missing value filling. Dimensional values;

[0048] Apply material compatibility boundary constraints after single-dimensional boundary constraints:

[0049] Define material sensitivity function Return material The following parameters Safety limit:

[0050] ;

[0051] After boundary processing, we get:

[0052] ;

[0053] For missing values, an intelligent missing value imputation strategy is adopted, the specific strategy is as follows:

[0054] First, determine the missing values:

[0055] Define indicator functions :

[0056] ;

[0057] Wherein, an output of 1 indicates the dimension of the parameter. Missing, unfilled, incorrectly formatted, or invalid values ​​require subsequent filling or correction using a priority model; an output of 0 indicates that the parameter dimension... The value has been correctly filled in by the user and is valid. No processing is required; simply retain the original value.

[0058] The logic of the priority model is as follows:

[0059] If users exist Historical preferences ,and exist:

[0060] but = This indicates that the indicator function output is 1 and that the user's historical preference for this dimension exists in the system.

[0061] If there are no user preferences, and Does not exist:

[0062] Then use the standard process parameters: This indicates that the indicator function output is 1 and there is no historical preference for this dimension for this user in the system.

[0063] like ,but This indicates that the indicator function outputs 0, meaning the user's input is correct and valid.

[0064] The complete vector after padding is:

[0065] ;

[0066] Preferably, the semantic parsing mathematical model transforms the user's free text into structured parameters. First, the user's free text... Preprocessing is performed to obtain word sequences. Specifically, it includes:

[0067] Through word segmentation function Obtain the word sequence ;

[0068] The target word sequence obtained after stop word filtering:

[0069] ;

[0070] Among them, word sequence This will serve as the input data for this filtering process; The difference operator represents a set, which removes all elements contained in the set from the set and keeps the remainder; Represents a set of stop words;

[0071] Perform synonym mapping on the filtered word sequence:

[0072] ;

[0073] in, It is a synonym mapping function.

[0074] Preferably, the preprocessed word sequence Perform parameter mapping to achieve the standard process parameter vector. User-filled process parameters Perform parameter dimension unification and coding standardization, specifically including:

[0075] Define rule dictionary ,in, Indicates keywords / regular expression patterns; This represents the target parameter dimension, corresponding to the process parameters: washing water temperature T, mechanical strength M, main wash time S, spin speed level D, permissible bleach type B, drying method R, high-temperature disinfection / mite removal H, fabric softener usage F, and process cost / energy consumption level C, compared with the standard process parameter vector. The meaning is consistent; Indicates the value of the mapping parameter;

[0076] The parameter mapping rules are as follows:

[0077] ;

[0078] in, Represents the dimension of the target parameter The set of candidate values; It is the union operator for sets; These are preprocessed user-defined free text words. Indicates a single candidate value A collection of single elements;

[0079] Define priority weights To resolve the conflict issue of multiple keywords mapping to the same dimension, Indicates safety / wear weight, Indicates the weight of cleaning effectiveness. This indicates the priority of cost considerations. Indicates the experience / odor weight, and the final value is:

[0080] ;

[0081] in, Represents the first element in the user process parameter vector. The final value of the dimension; This represents the parameter maximization operator; Indicates candidate values; Indicates candidate value Priority weights.

[0082] Preferably, the process knowledge base data structure includes:

[0083] Define knowledge base entries, with each knowledge base record being a triple. ;

[0084] in, Indicates the first Each record contains the feature vector of the clothing to be adapted. Indicates the first The standard process parameter vector corresponding to each record; Indicates the priority of the basic process;

[0085] Establish a secondary retrieval index for the knowledge base, starting with the clothing category in the primary index. Quickly locate the main category, then enter the secondary index: fabric material. To narrow down the search scope;

[0086] The clothing feature vector input by the user With the knowledge base The encoding format is uniform, and the matching rules for each component are consistent.

[0087] Based on the secondary retrieval index, the knowledge base is traversed to select the candidate set K that meets the following constraints:

[0088] ;

[0089] in, This represents a category compatibility function that returns a set of categories compatible with the target category. This indicates a parent color function that returns the parent color class of the target color.

[0090] For each record in candidate set K Calculate the matching score :

[0091] ;

[0092] in, Represents user feature vector The One component; It is a similarity function with a value range of [0,1]; completely identical is 1, compatible is 0.5, and incompatible is 0. The weights of each feature classification satisfy the following conditions: ;

[0093] Select the record with the highest score from candidate set K. The corresponding process parameter vector is the search result:

[0094] ;

[0095] ;

[0096] in, Indicates the optimal entry The corresponding standard process parameter vector, i.e., the first one retrieved from the knowledge base. The process parameters of each record are assigned to the final result. ;

[0097] Feedback data from search results is used to optimize weights. and similarity function The formula is as follows:

[0098] ;

[0099] in, Indicates the first After round of iteration, the first Updated weights for each feature dimension; Indicates to The gradient operator for partial derivatives, representing the loss function. Follow The direction and rate of change; This indicates that the loss function applies to the current weights. The gradient indicates how to reduce the loss. The direction and size that should be adjusted It is a loss function that reflects the complaint rate caused by mismatched processes; This represents the learning rate.

[0100] Preferably, if the candidate set K is empty, the candidate set is re-filtered, and the user is prompted that there is no standard process that completely matches the garment, and a compatible process has been adopted. It is recommended to confirm the material information.

[0101] The preferred design for the difference calculation model is as follows:

[0102] For numerical / rank parameters, their one-dimensional variability Represented by normalized absolute difference, i.e.:

[0103] ;

[0104] in, The first element representing the standard process parameter vector Dimensional values;

[0105] For enumerated / Boolean parameters, the one-dimensional variability Represented by a custom difference matrix, namely:

[0106] ;

[0107] in, Indicates the first A custom difference matrix function with dimension parameter that maps semantic differences to [0,1] values ​​based on enumeration / Boolean types;

[0108] Then the overall difference ;

[0109] in, Indicates the first The weighting coefficients of the dimension parameter satisfy the following conditions: A higher weight indicates that the parameter has a greater impact on washing safety / performance.

[0110] Preferably, based on the overall degree of difference Risk level classification:

[0111] If , it indicates low risk;

[0112] If it is , then it indicates medium risk;

[0113] This indicates high risk;

[0114] in, , This indicates the threshold for classifying risk levels;

[0115] Define risk type set The risk type identification and determination rules are as follows:

[0116] Security risks: , or ;

[0117] This indicates that at least one safety parameter has exceeded the limit. Safety parameters include water temperature. bleaching type Drying method Dehydration speed ; Indicates the dimension of this parameter The physical limit maximum value;

[0118] Quality risks: and ( );

[0119] This indicates that the overall difference in multiple effect parameters exceeds the standard and the user-set parameters are weaker than the standard process. Effect parameters include mechanical strength. Main washing time Main wash temperature ;

[0120] Waste of cost: And there is no safety / mass gain;

[0121] Experience bias: And users explicitly mentioned that it was good;

[0122] in, This indicates that the discrepancies in the user experience parameters exceeded the acceptable limits, and that the user had explicitly requested something that was not met. These user experience parameters include the amount of fabric softener used. High temperature treatment ; , , , All are type recognition thresholds.

[0123] Preferably, a conflict rule verification model is defined. Iterate through all rules in order, if Then record the conflict type. In addition to the design parameters, a severity level is set for the conflict.

[0124] The present invention has the following advantages:

[0125] 1. This invention uses a lightweight model that combines text parsing and rules to map user-filled requirements into a structured parameter vector, and performs difference calculations with standard process parameters in the same space, thus realizing the transformation from free text to quantified differences.

[0126] 2. This invention not only calculates the overall difference, but also evaluates the water temperature, mechanical strength, dehydration, drying, bleaching and other dimensions separately, and proposes multiple risk types such as safety risk, quality risk, cost waste and experience deviation. The multi-dimensional parameters are jointly evaluated to distinguish the risk types and provide refined support for platform decision-making.

[0127] 3. By generating explanatory text and several alternative process solutions through templates and rules, the workload of manual customer service in interpreting notes for each order is greatly reduced. Explanations and process suggestions are automatically generated, reducing labor costs and improving order processing efficiency and consistency.

[0128] 4. This invention is based on standard parameter vectors, rule engines, and statistical learning methods. It can be easily integrated into existing Internet laundry platforms or traditional laundry management systems, and can be flexibly expanded according to the standard process libraries of different enterprises, making it easy to integrate and expand. Detailed Implementation

[0129] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0130] This invention proposes a method for verifying the difference between user-filled washing requirements and standard processes. The method includes: based on a process knowledge base, analyzing the feature vectors of clothing... The corresponding standard process parameter vector is obtained through the standard process retrieval algorithm. A secondary retrieval index is constructed in the knowledge base, and then the process knowledge base is traversed to obtain a candidate set that meets the constraints and the record with the highest score is selected to obtain the recommended process and safety boundary.

[0131] The clothing feature vector G includes user-selected options and user-filled process parameters. ; Standard process parameter vectors are analyzed through semantic parsing mathematical models. User-filled process parameters Perform parameter dimension unification and coding standardization; calculate the single-dimensional difference degree one by one through the difference degree calculation model. Combining the weights of each process parameter dimension To obtain the overall difference It combines thresholds to classify risk levels and identify risk types; at the same time, it makes hard conflict judgments through a conflict rule verification model to identify safety risks, quality risks, cost waste and experience deviations, and provides a basis for judgment to generate difference results.

[0132] Based on the discrepancy results, an explanation template and several alternative process solutions are generated to serve the user. The final solution and subsequent complaints and rewashing situations are recorded. Weights and thresholds are adjusted periodically using historical data, forming a closed-loop discrepancy verification method driven by knowledge base, rules, and data. The explanation template is generated using the native functionality of the slot filling algorithm.

[0133] As an example of the present invention, a clothing feature vector is defined. ;

[0134] in, The clothing category is indicated, such as "shirt", "down jacket", "sweater", which comes from user selection or store entry;

[0135] This indicates the fabric material, such as "cotton," "polyester," "wool," "silk," etc., which is selected by the user when placing an order or obtained by scanning a barcode / taking a photo to identify the label at the store.

[0136] Indicates the color type, such as "light", "dark", "pure white", etc.

[0137] This indicates information about filling and accessories, such as "down filling", "leather patchwork", "metal zipper", etc.

[0138] This represents a set of washing label symbols, a binary vector obtained through photo recognition or manual input. For example, "Do not bleach" = 1, "Do not tumble dry" = 1, and so on. One tag;

[0139] Standard process parameter vectors are obtained by predefining the process knowledge base and retrieving them based on clothing feature vectors. ;

[0140] in, The standard washing water temperature is indicated in °C and is derived from the process knowledge base and washing labels. The standard mechanical strength grade (e.g., 1–5) is derived from the process knowledge base; This indicates the standard main wash time, in minutes. Indicates the standard dehydration speed level (e.g., 0 = no dehydration, 1 = low speed, 2 = medium speed, 3 = high speed). Indicates bleaching permission symbols (0 = prohibited, 1 = oxygen bleaching allowed, 2 = chlorine bleaching allowed, etc.). The code indicates the standard drying method (0 = air drying, 1 = low temperature drying, 2 = medium temperature drying, 3 = high temperature drying). Indicates whether it includes a high-temperature disinfection / mite removal function; Indicates the fabric softener usage mode (0 = no, 1 = standard amount, 2 = extra). This indicates the recommended unit cost or energy consumption level, which facilitates subsequent cost difference determination;

[0141] User-inputted process parameters ;

[0142] in, This indicates the user's requested washing water temperature; This indicates the user's required mechanical strength grade; This indicates the user's requested main wash time; This indicates the user's required dehydration speed level; This indicates the user's bleaching request; This indicates the user's requested drying method; This indicates the user's requirement for high-temperature disinfection / mite removal; This indicates the user's preference for fabric softener; This indicates the user's expected unit cost or energy consumption level; making and Dimensional alignment and consistent value selection rules are maintained; if no explicit user requirements are specified, the standard process will be executed by default. ;in, Indicates the parameter dimension;

[0143] The rules for selecting values ​​are as follows:

[0144] First, unify data types:

[0145] Numerical data: washing water temperature T, main wash time S, process cost / energy consumption level C, all stored as floating-point numbers;

[0146] Graded type: Mechanical strength M, dehydration speed grade D, fabric softener usage F, unified integer code;

[0147] Bleaching type B and drying method R are allowed, and a unified integer encoding mapping table is used.

[0148] Boolean type: High-temperature disinfection / mite removal H, unified integer encoding;

[0149] The boundary value handling mechanism is as follows:

[0150] Single-dimensional boundary constraints:

[0151] For numerical / rank parameters:

[0152] ;

[0153] For each parameter dimension Define its safe value range Made of fabric material Set of washing label symbols Decision; among them, For the first User values ​​for each parameter dimension;

[0154] For enumeration / Boolean parameters:

[0155] ;

[0156] in, This is the set of legal values ​​for this dimension; For the first Standard values ​​for each parameter dimension; This represents the first value in the user process parameter vector after boundary handling and missing value filling. Dimensional values;

[0157] Apply material compatibility boundary constraints after single-dimensional boundary constraints:

[0158] Define material sensitivity function Return material The following parameters Safety limit:

[0159] ;

[0160] After boundary processing, we get:

[0161] ;

[0162] For missing values, an intelligent missing value imputation strategy is adopted, the specific strategy is as follows:

[0163] First, determine the missing values:

[0164] Define indicator functions :

[0165] ;

[0166] Where an output of 1 indicates the dimension of the parameter. Missing, unfilled, incorrectly formatted, or invalid values ​​require subsequent filling or correction using a priority model; an output of 0 indicates that the parameter dimension... The value has been correctly filled in by the user and is valid. No processing is required; simply retain the original value.

[0167] The logic of the priority model is as follows:

[0168] If users exist Historical preferences ,and exist:

[0169] but = This indicates that the indicator function output is 1 and that the user's historical preference for this dimension exists in the system.

[0170] If there are no user preferences, and Does not exist:

[0171] Then use the standard process parameters: This indicates that the indicator function output is 1 and there is no historical preference for this dimension for this user in the system.

[0172] like ,but This indicates that the indicator function outputs 0, meaning the user's input is correct and valid.

[0173] The complete vector after padding is:

[0174] ;

[0175] As an example of the present invention, the semantic parsing mathematical model transforms user-defined free text into structured parameters. First, the user-defined free text... Preprocessing is performed to obtain word sequences. Specifically, it includes:

[0176] Through word segmentation function Obtain the word sequence As shown in Table 1;

[0177] Table 1

[0178]

[0179] The target word sequence obtained after stop word filtering:

[0180] ;

[0181] Among them, word sequence This will serve as the input data for this filtering process; The difference operator represents a set, which removes all elements contained in the set from the set and keeps the remainder; Represents a set of stop words;

[0182] Perform synonym mapping on the filtered word sequence:

[0183] ;

[0184] in, It is a synonym mapping function.

[0185] As an example of the present invention, the preprocessed word sequence Perform parameter mapping to achieve the standard process parameter vector. User-filled process parameters Perform parameter dimension unification and coding standardization, specifically including:

[0186] Define rule dictionary ,in, Indicates keywords / regular expression patterns; This represents the target parameter dimension, corresponding to the process parameters: washing water temperature T, mechanical strength M, main wash time S, spin speed level D, permissible bleach type B, drying method R, high-temperature disinfection / mite removal H, fabric softener usage F, and process cost / energy consumption level C, compared with the standard process parameter vector. The meaning is consistent; Indicates the value of the mapping parameter (e.g., (This indicates high-temperature mite removal).

[0187] The parameter mapping rules are as follows:

[0188] ;

[0189] in, Represents the dimension of the target parameter The set of candidate values; It is the union operator for sets; These are preprocessed user-defined free text words. Indicates a single candidate value The single-element set formed is shown in Table 2;

[0190] Table 2

[0191] Keywords / Regular Expressions Parameter Dimensions Parameter value Cold water / Low temperature water temperature ≤30℃ 40 degrees water temperature 40℃ Don't use too much force. Mechanical strength Low Powerful stain remover Mechanical strength high Do not dehydrate Dehydration Non-dehydration High-temperature drying drying high temperature Do not bleach bleach prohibit Add more fabric softener fabric softener strengthen mite removal / disinfection Additional features Enable

[0192] Define priority weights To resolve the conflict issue of multiple keywords mapping to the same dimension, Indicates safety / wear weight, Indicates the weight of cleaning effectiveness. This indicates the priority of cost considerations. Indicates the experience / odor weight, and the final value is:

[0193] ;

[0194] in, Represents the first element in the user process parameter vector. The final value of the dimension; This represents the parameter maximization operator; Indicates candidate values; Indicates candidate value Priority weights; for example, priority weights set up This indicates that you should not wash vigorously; priority weight. set up 4 indicates that it is clean and should be taken as the final product. As shown in Table 3;

[0195] Table 3

[0196] Keyword types Priority Safety / wear 3 Cleaning effect 2 Experience / Scent 1

[0197] As an example of the present invention, the process knowledge base data structure includes:

[0198] Define knowledge base entries, with each knowledge base record being a triple. ;

[0199] in, Indicates the first Each record contains the feature vector of the clothing to be adapted. Indicates the first The standard process parameter vector corresponding to each record; Indicates the priority of the basic process;

[0200] Establish a secondary retrieval index for the knowledge base, starting with the clothing category in the primary index. Quickly locate the main category, then enter the secondary index: fabric material. To narrow down the search scope;

[0201] The clothing feature vector input by the user With the knowledge base The encoding format is uniform, and the matching rules for each component are consistent, as shown in Table 4:

[0202] Table 4

[0203]

[0204] Based on the secondary retrieval index, the knowledge base is traversed to select the candidate set K that meets the following constraints:

[0205] ;

[0206] in, This represents a category compatibility function that returns a set of categories compatible with the target category. This indicates a parent color function that returns the parent color class of the target color.

[0207] For each record in candidate set K Calculate the matching score :

[0208] ;

[0209] in, Represents user feature vector The One component; It is a similarity function with a value range of [0,1]; completely identical is 1, compatible is 0.5, and incompatible is 0. The weights of each feature classification satisfy the following conditions: ;

[0210] Select the record with the highest score from candidate set K. The corresponding process parameter vector is the search result:

[0211] ;

[0212] ;

[0213] in, Indicates the optimal entry The corresponding standard process parameter vector, i.e., the first one retrieved from the knowledge base. The process parameters of each record are assigned to the final result. ;

[0214] Feedback data from search results is used to optimize weights. and similarity function The formula is as follows:

[0215] ;

[0216] in, Indicates the first After round of iteration, the first Updated weights for each feature dimension; Indicates to The gradient operator for partial derivatives, representing the loss function. Follow The direction and rate of change; This indicates that the loss function applies to the current weights. The gradient; It is a loss function that reflects the complaint rate caused by mismatched processes; This represents the learning rate, set to 0.01.

[0217] As an example of the present invention, if the candidate set K is empty, the candidate set is re-filtered, and the user is prompted that there is no standard process that completely matches the garment, and a compatible process has been adopted. It is recommended to confirm the material information.

[0218] As an example of the present invention, the difference calculation model is designed as follows:

[0219] For numerical / rank parameters, their one-dimensional variability Represented by normalized absolute difference, i.e.:

[0220] ;

[0221] in, The first element representing the standard process parameter vector Dimensional values;

[0222] For enumerated / Boolean parameters, the one-dimensional variability Represented by a custom difference matrix, namely:

[0223] ;

[0224] in, Indicates the first A custom difference matrix function with dimension parameter that maps semantic differences to [0,1] values ​​based on enumeration / Boolean types;

[0225] Then the overall difference ;

[0226] in, Indicates the first The weighting coefficients of the dimension parameter satisfy the following conditions: .

[0227] As an example of the present invention, based on the comprehensive difference degree Risk level classification:

[0228] If , it indicates low risk;

[0229] If it is , then it indicates medium risk;

[0230] This indicates high risk;

[0231] in, , This indicates the threshold for classifying risk levels;

[0232] Define risk type set The risk type identification and determination rules are as follows:

[0233] Security risks: , or ( );

[0234] This indicates that at least one safety parameter has exceeded the limit. Safety parameters include water temperature. bleaching type Drying method Dehydration speed ; Indicates the dimension of this parameter The physical limit maximum value;

[0235] Quality risks: and ( );

[0236] This indicates that the overall difference in multiple effect parameters exceeds the standard and the user-set parameters are weaker than the standard process. Effect parameters include mechanical strength. Main washing time Main wash temperature ;

[0237] Waste of cost: And there is no safety / mass gain ( );

[0238] Experience bias: And users explicitly mentioned that it was good;

[0239] in, This indicates that the discrepancies in the user experience parameters exceeded the acceptable limits, and that the user had explicitly requested something that was not met. These user experience parameters include the amount of fabric softener used. High temperature treatment ; , , , All are type recognition thresholds.

[0240] As an example of the present invention, a conflict rule verification model is defined. Iterate through all rules in order, if Then record the conflict type. In addition to the design parameters, a severity level is set for the conflict.

[0241] In actual system implementation, a master table combined with a related table pattern is adopted to ensure that clothing features and business information are decoupled but traceable. User ID and order ID are used as related metadata and clothing feature vectors. Binding storage (rather than integrating into clothing feature vectors) This forms a link between users, orders, garment characteristics, and processes, used to record the final solution and subsequent complaints and rewashing situations. This data is then used to adjust weights and thresholds based on historical data, as detailed in Table 5.

[0242] Table 5

[0243]

[0244] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for verifying the difference between user-defined washing requirements and standard processes, characterized in that: The methods include: Based on the process knowledge base, the feature vectors of clothing The corresponding standard process parameter vector is obtained through the standard process retrieval algorithm. A secondary retrieval index is constructed in the knowledge base, and then the process knowledge base is traversed to obtain a candidate set that meets the constraints and the record with the highest score is selected to obtain the recommended process and safety boundary. The clothing feature vector G includes user-selected options and user-filled process parameters. ; Standard process parameter vectors are analyzed through semantic parsing mathematical models. User-filled process parameters Perform parameter dimension unification and coding standardization; calculate the single-dimensional difference degree one by one through the difference degree calculation model. Combining the weights of each process parameter dimension To obtain the overall difference It combines thresholds to classify risk levels and identify risk types; at the same time, it makes hard conflict judgments through a conflict rule verification model to identify safety risks, quality risks, cost waste and experience deviations, and provides a basis for judgment to generate difference results; Based on the discrepancy results, an explanation template and several alternative process solutions are generated to serve the user. The final solution and subsequent complaints and rewashing situations are recorded. Weights and thresholds are adjusted regularly using historical data to form a closed-loop discrepancy verification method driven by knowledge base, rules and data. Specifically, the parameter dimension unification and encoding standardization are as follows: Semantic parsing mathematical models transform user-defined free text into structured parameters. First, the user-defined free text... Preprocessing is performed to obtain the word sequence. ; For the preprocessed word sequence Perform parameter mapping to achieve the standard process parameter vector. User-filled process parameters Perform parameter dimension unification and coding standardization, specifically including: Define rule dictionary ,in, Indicates keywords / regular expression patterns; This represents the target parameter dimension, corresponding to the process parameters: washing water temperature T, mechanical strength M, main wash time S, spin speed level D, permissible bleach type B, drying method R, high-temperature disinfection / mite removal H, fabric softener usage F, and process cost / energy consumption level C, compared with the standard process parameter vector. The meaning is consistent; Indicates the value of the mapping parameter; The parameter mapping rules are as follows: ； in, Represents the dimension of the target parameter The set of candidate values; It is the union operator for sets; It is a single word of user-free text after preprocessing; Indicates a single candidate value A collection of single elements; Define priority weights To resolve the conflict issue of multiple keywords mapping to the same dimension, Indicates safety / wear weight, Indicates the weight of cleaning effectiveness. This indicates the priority of cost considerations. Indicates the experience / odor weight, and the final value is: ； in, Represents the first element in the user process parameter vector. The final value of the dimension; This represents the parameter maximization operator; Indicates candidate values; Indicates candidate value Priority weights.

2. The method for verifying the difference between user-filled washing requirements and standard processes according to claim 1, characterized in that: Define clothing feature vector ; in, The clothing category is indicated by user selection or store entry. Indicates the fabric material; Indicates the color type; Indicates information about filling and auxiliary materials; This represents a set of washing label symbols, which is a binary vector obtained through photo recognition or manual input. It contains a total of [number missing]. One tag; Standard process parameter vectors are obtained by predefining the process knowledge base and retrieving them based on clothing feature vectors. ; in, The standard washing water temperature is indicated in °C and is derived from the process knowledge base and washing labels. The standard mechanical strength grade is derived from the process knowledge base; This indicates the standard main wash time, in minutes. Indicates the standard dehydration speed rating; Indicates a bleaching permit mark; Indicates the standard drying method code; Indicates whether it includes a high-temperature disinfection / mite removal function; Indicates the method of using fabric softener; This indicates the recommended unit cost or energy consumption level; User-inputted process parameters ; in, This indicates the user's requested washing water temperature; This indicates the user's required mechanical strength grade; This indicates the user's requested main wash time; This indicates the user's required dehydration speed level; This indicates the user's bleaching request; This indicates the user's requested drying method; This indicates the user's requirement for high-temperature disinfection / mite removal; This indicates the user's preference for fabric softener; This indicates the user's expected unit cost or energy consumption level; making and Dimensional alignment and consistent value selection rules are maintained; if no explicit user requirements are specified, the standard process will be executed by default. ;in, Indicates the parameter dimension; The rules for selecting values ​​are as follows: First, unify data types: Numerical data: Washing water temperature T, main wash time S, process cost / energy consumption level C, all stored as floating-point numbers; Graded type: Mechanical strength M, dehydration speed grade D, fabric softener use F, unified integer code; Bleaching type B and drying method R are allowed, and a unified integer encoding mapping table is used. Boolean type: High-temperature disinfection / mite removal H, unified integer encoding; The boundary value handling mechanism is as follows: Single-dimensional boundary constraints: For numerical / rank parameters: ； For each parameter dimension Define its safe value range Made of fabric material Set of washing label symbols Decision; among them, For the first User values ​​for each parameter dimension; For enumeration / Boolean parameters: ; in, This is the set of legal values ​​for this dimension; For the first Standard values ​​for each parameter dimension; This represents the first value in the user process parameter vector after boundary handling and missing value filling. Dimensional values; Apply material compatibility boundary constraints after single-dimensional boundary constraints: Define material sensitivity function Return material The following parameters Safety limit: ; After boundary processing, we get: ; For missing values, an intelligent missing value imputation strategy is adopted, the specific strategy is as follows: First, determine the missing values: Define indicator functions : ； Wherein, an output of 1 indicates the dimension of the parameter. Missing, unfilled, incorrectly formatted, or invalid values ​​require subsequent filling or correction using a priority model; an output of 0 indicates that the parameter dimension... The value has been correctly filled in by the user and is valid. No processing is required; simply retain the original value. The logic of the priority model is as follows: If users exist Historical preferences ,and exist: but = This indicates that the indicator function output is 1 and that the user's historical preference for this dimension exists in the system. If there are no user preferences, and Does not exist: Then use the standard process parameters: This indicates that the indicator function output is 1 and there is no historical preference for this dimension for this user in the system. like ,but This indicates that the indicator function outputs 0, meaning the user's input is correct and valid. The complete vector after padding is: 。 3. The method for verifying the difference between user-filled washing requirements and standard processes according to claim 1, characterized in that: Semantic parsing mathematical models transform user-defined free text into structured parameters. First, the user-defined free text... Preprocessing is performed to obtain the word sequence. Specifically, it includes: Through word segmentation function Obtain the word sequence ; The target word sequence obtained after stop word filtering: ; Among them, word sequence This will serve as the input data for this filtering process; The difference operator represents a set, which removes all elements contained in the set from the set and keeps the remainder; Represents a set of stop words; Perform synonym mapping on the filtered word sequence: ; in, It is a synonym mapping function.

4. The method for verifying the difference between user-filled washing requirements and standard processes according to claim 3, characterized in that: The data structure of the process knowledge base includes: Define knowledge base entries, with each knowledge base record being a triple. ; in, Indicates the first Each record contains the feature vector of the clothing to be adapted. Indicates the first The standard process parameter vector corresponding to each record; Indicates the priority of the basic process; To build a secondary retrieval index for the knowledge base, first use the primary index for clothing categories. Quickly locate the main category, then enter the secondary index: fabric material. To narrow down the search scope; The clothing feature vector input by the user With the knowledge base The encoding format is uniform, and the matching rules for each component are consistent. Based on the secondary retrieval index, the knowledge base is traversed to select the candidate set K that meets the following constraints: ； in, This represents a category compatibility function that returns a set of categories compatible with the target category. This indicates a parent color function that returns the parent color class of the target color. For each record in candidate set K Calculate the matching score : ； in, Represents user feature vector The One component; It is a similarity function with a value range of [0,1]; completely identical is 1, compatible is 0.5, and incompatible is 0. The weights of each feature classification satisfy the following conditions: ; Select the record with the highest score from candidate set K. The corresponding process parameter vector is the search result: ； ； in, Indicates the optimal entry The corresponding standard process parameter vector, i.e., the first one retrieved from the knowledge base. The process parameters of each record are assigned to the final result. ; Feedback data from search results is used to optimize weights. and similarity function The formula is as follows: ； in, Indicates the first After round of iteration, the first Updated weights for each feature dimension; Indicates to The gradient operator for partial derivatives, representing the loss function. Follow The direction and rate of change; This indicates that the loss function applies to the current weights. The gradient; It is a loss function that reflects the complaint rate caused by mismatched processes; This represents the learning rate.

5. The method for verifying the difference between user-filled washing requirements and standard processes according to claim 4, characterized in that: If candidate set K is empty, the candidate set will be filtered again, and the user will be prompted that there is no standard process that completely matches the garment and that a compatible process has been adopted. It is recommended to confirm the material information.

6. The method for verifying the difference between user-filled washing requirements and standard processes according to claim 5, characterized in that: The difference calculation model is designed as follows: For numerical / rank parameters, their one-dimensional variability Represented by normalized absolute difference, i.e.: ； in, The first element representing the standard process parameter vector Dimensional values; For enumerated / Boolean parameters, the one-dimensional variability Represented by a custom difference matrix, namely: ； in, Indicates the first A custom difference matrix function with dimension parameter that maps semantic differences to [0,1] values ​​based on enumeration / Boolean types; Then the overall difference ; in, Indicates the first The weighting coefficients of the dimension parameter satisfy the following conditions: .

7. The method for verifying the difference between user-filled washing requirements and standard processes according to claim 6, characterized in that: Based on overall difference Risk level classification: If , it indicates low risk; If it is , then it indicates medium risk; This indicates high risk; in, , This indicates the threshold for classifying risk levels; Define risk type set The risk type identification and determination rules are as follows: Security risks: , or ; This indicates that at least one safety parameter has exceeded the limit. Safety parameters include water temperature. bleaching type Drying method Dehydration speed ; Indicates the dimension of this parameter The physical limit maximum value; Quality risks: and ; This indicates that the overall difference in multiple effect parameters exceeds the standard and the user-set parameters are weaker than the standard process. Effect parameters include mechanical strength. Main washing time Main wash temperature ; Waste of cost: And there is no safety / mass gain; Experience bias: And users explicitly mentioned that it was good; in, This indicates that the discrepancies in the user experience parameters exceeded the acceptable limits, and that the user had explicitly requested something that was not met. These user experience parameters include the amount of fabric softener used. High temperature treatment ; , , , All are type recognition thresholds.

8. The method for verifying the difference between user-filled washing requirements and standard processes according to claim 7, characterized in that: Define a conflict rule verification model traverse all rules in order, if Then record the conflict type. In addition to the design parameters, a severity level is set for the conflict.

Images