A garment data processing method and device based on an enterprise-level operating system

Abstract

The embodiment of the application discloses a kind of clothing data processing method and equipment based on enterprise-level operating system, belong to clothing data processing technical field, solve clothing industry data entry link, due to the higher error rate of manual operation, affect the accuracy and completeness of industry data problem.The entity data recognition and format uniformity of clothing basic data are carried out by prepositioning multimodal pre-training model;Clothing basic data is input into different data modules, and the knowledge graph corresponding to entity data and different data modules respectively, to detect error data for clothing basic data;The repair value prediction of detected error data is carried out by data prediction filling model, and the clothing data to be fused is generated by repaired data;The feature vector corresponding to the clothing data to be fused and historical clothing data respectively is associated degree processing, and constructs clothing association relationship network;Based on clothing association relationship network, the clothing data to be fused is fused to the corresponding position of system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of clothing data processing, and particularly relates to a clothing data processing method and device based on an enterprise-level operating system. BACKGROUND

[0002] In the clothing industry, there are many complex characteristics in business operation. On the one hand, the inventory holding units of clothing products are extremely complex, covering various combinations of different styles, colors, sizes, and the like, and the bill of materials level is numerous, involving many links from raw materials to finished products. On the other hand, the traditional ERP system faces great challenges in the application of clothing enterprises. For example, more than 2000 basic parameters need to be configured when the traditional ERP system is initialized, which makes the enterprise spend an average of 3 to 6 months to deploy, greatly consuming the human, material and time costs of the enterprise.

[0003] In addition, the process and procedure in the clothing production process have highly specialized characteristics, but there is currently a lack of unified and standardized standard system in the industry. This situation directly leads to errors in manual operation in the data entry link, seriously affecting the accuracy and integrity of the data in the clothing industry. SUMMARY

[0004] The embodiment of the present application provides a clothing data processing method and device based on an enterprise-level operating system, which is used to solve the technical problem that there is currently a lack of unified and standardized standard system in the clothing industry, which leads to errors in manual operation in the data entry link, seriously affecting the accuracy and integrity of the data in the clothing industry.

[0005] The embodiment of the present application adopts the following technical scheme:

[0006] The embodiment of the present application provides a clothing data processing method based on an enterprise-level operating system. The method comprises inputting obtained clothing basic data into a preset multi-modal pre-training model to perform entity data recognition and format unification on the clothing basic data through the preset multi-modal pre-training model; based on the data type of the clothing basic data, inputting the clothing basic data into different data modules, and performing error data detection on the clothing basic data through the knowledge graph corresponding to the entity data and the different data modules; performing repair value prediction on the detected error data through a data prediction filling model, generating corresponding to-be-fused clothing data of each data module through the repaired data; performing correlation degree processing on the feature vectors corresponding to the to-be-fused clothing data and the feature vectors corresponding to historical clothing data, and constructing a clothing correlation relationship network; and based on the clothing correlation relationship network, fusing the to-be-fused clothing data to the corresponding position of the system and performing data publishing.

[0007] The embodiment of the application can effectively correct the confusion caused by non-uniform data formats, entity recognition errors, etc. by pre-setting a multi-modal pre-training model for entity data recognition and format unification of clothing basic data. In combination with a knowledge graph for error data detection, and using a data prediction filling model to repair error data, the accuracy of the data is comprehensively ensured, and the influence of manual input errors on subsequent processes is reduced. Secondly, the clothing basic data is input into different data modules for parallel processing according to the data type, and the knowledge graph is used to quickly locate errors, which saves time compared with the traditional single processing mode. At the same time, the model-based error repair and correlation relationship construction have high automation, avoiding a large amount of manual operation, and further improving the data processing efficiency. In addition, by constructing a clothing correlation relationship network, the to-be-fused clothing data is reasonably fused to the corresponding position of the system and published, which can enable the enterprise to clearly grasp the correlation between the clothing data, and the data of each link is connected from raw material procurement, production process to finished product sales, which assists the enterprise to make more scientific decisions and improves the smoothness of overall operation management.

[0008] In an implementation manner of the application, the obtained clothing basic data is input into a pre-set multi-modal pre-training model to perform entity data recognition and format unification on the clothing basic data through the pre-set multi-modal pre-training model. Specifically, the pre-set OCR component is used to extract text from unstructured data in the clothing basic data, and the extracted text and the clothing basic data are input into the pre-set multi-modal pre-training model; wherein the clothing basic data includes full-process data of data acquisition, data processing and data storage; the pre-set multi-modal pre-training model is used to perform entity data recognition on the clothing basic data; based on the data type corresponding to the entity data, a corresponding pre-set data format template is determined to perform format unification on the clothing basic data through the pre-set data format template.

[0009] In an implementation manner of the application, after the clothing basic data is recognized by the pre-set multi-modal pre-training model, the method further includes: performing multi-dimensional data detection on the clothing basic data based on a pre-set data detection rule; performing weighted processing on the multi-dimensional data detection result to determine the data quality level corresponding to the clothing basic data based on the weighted processing result; and marking abnormal data in the clothing basic data in different forms based on different data quality levels to trigger corresponding data processing nodes based on the marking form.

[0010] In an implementation manner of the present application, the error data detection of the clothing basic data is performed through the entity data and the knowledge graph corresponding to different data modules, and specifically includes: determining the logical relationship between different entity data based on the knowledge graph; wherein the logical relationship at least includes one of production relationship, design relationship and quality relationship; performing business path detection on the logical relationship between the clothing basic data based on the predetermined clothing business path; comparing the entity data in the clothing basic data with the entity relationship in the preset knowledge graph to perform data logic detection; determining the corresponding historical data based on different types of entity data, comparing the entity data with the corresponding historical data to determine the data difference; determining the corresponding detection error threshold based on different types of entity data, and performing error value detection based on the detection error threshold and the data difference; and realizing the error detection of the clothing basic data through the business path detection, the data logic detection and the error value detection.

[0011] In an implementation manner of the present application, the error data detected is repaired and predicted through a data prediction filling model, and specifically includes: determining the associated clothing data corresponding to the error data in the clothing basic data, and extracting the keywords of the associated clothing data; converting the text data in the keywords into a low-dimensional dense vector, and scaling the numerical data in the keywords to a preset numerical section; sorting and combining the converted text data and the scaled numerical data to construct a model input sequence; inputting the model input sequence into the data prediction filling model, and obtaining a predicted repair value through the operation of the multi-layer Transformer layer in the data prediction filling model.

[0012] In an implementation manner of the present application, the error data detected is repaired and predicted through a data prediction filling model, and the method further includes: in the case that the error data has multiple data source conflicts, determining the information entropy corresponding to the multiple data sources in different dimensions respectively; determining the weight of the multiple data sources based on the information entropy; determining the optimal data source data based on the weight value, and predicting the repair value of the detected error data based on the optimal data source data.

[0013] In an implementation form of the present application, the feature vector corresponding to the to-be-fused clothing data is associated with the feature vector corresponding to the historical clothing data, and a clothing association relationship network is constructed, specifically including: based on the pre-defined data relationship in the knowledge graph, an association entity graph is constructed, and a plurality of association entity pairs are determined based on the association entity graph; the feature vector corresponding to the to-be-fused clothing data is associated with the feature vector corresponding to the historical clothing data, and based on the association degree, a reference association pair is determined; the reference association pair existing in the association entity graph is labeled, and the reference association pair not existing in the association entity graph is connected to construct an association data network; in the historical clothing data, the co-occurrence frequency corresponding to each of the plurality of association data pairs in the association data network is determined, and the association weight corresponding to each of the plurality of association data pairs is determined based on the co-occurrence frequency; based on the association degree and the association weight, the comprehensive association degree value corresponding to each of the plurality of association data is determined, and based on the comprehensive association degree value and the association data network, the clothing association relationship network is constructed.

[0014] In an implementation form of the present application, the association weight corresponding to each of the plurality of association data pairs is determined based on the co-occurrence frequency, specifically including: based on the function:

[0015]

[0016] The association weight corresponding to each of the plurality of association data pairs is determined; wherein w i is the association weight; n is the number of association data pairs; i is the serial number of the association data pair; f i is the co-occurrence frequency; w i is the association weight; a is the penalty factor; γ is the error; is the penalty term.

[0017] In an implementation form of the present application, based on the clothing association relationship network, the to-be-fused clothing data is fused into the corresponding position of the system, specifically including: based on the clothing association relationship network, the association historical data corresponding to the to-be-fused clothing data is determined; the association mapping between the to-be-fused clothing data and the association historical data is constructed, and the to-be-fused clothing data is filled into the corresponding position; after the to-be-fused clothing data is filled into the corresponding position, the method further includes: the to-be-fused data is labeled with data information in the corresponding position; wherein the data information at least includes data source and time stamp; based on the labeled data information, the relationship iteration chain corresponding to the to-be-fused data is constructed.

[0018] The embodiment of the application provides a kind of clothing data processing equipment based on enterprise operating system, comprising: at least one processor;And, with at least one processor communication connection's memory;Wherein, memory stores the instruction that can be executed by at least one processor, instruction is executed by at least one processor, to enable at least one processor can: the clothing basic data of acquisition is input to preset multi-modal pre-training model, to pass through preset multi-modal pre-training model to clothing basic data and format uniformity are carried out entity data identification;Based on the data type of clothing basic data, clothing basic data is input into different data module, and through entity data and the knowledge graph corresponding to different data module, clothing basic data is carried out error data detection;The repair value prediction of error data detected is carried out by data prediction filling model, and the feature vector corresponding to the data of each data module corresponding to the data of each data module is generated by the data after repair;The feature vector corresponding to the data to be fused clothing data is associated with the feature vector corresponding to the historical clothing data, and constructs clothing association relationship network;Based on clothing association relationship network, the data to be fused clothing data is fused to the corresponding position of system, and data publishing is carried out.

[0019] The above at least one technical solution adopted by the embodiment of the application can achieve the following beneficial effects: the embodiment of the application carries out entity data identification and format uniformity on clothing basic data through preset multi-modal pre-training model, which can effectively correct the confusion caused by non-uniform data format, entity recognition error, etc.Combined with knowledge graph for error data detection, and using data prediction filling model to repair error data, the accuracy of data is guaranteed from all aspects, and the influence of manual input error on subsequent process is reduced.Secondly, the embodiment of the application inputs clothing basic data into different data modules according to data type and processes in parallel, which saves time compared with traditional single processing mode.Meanwhile, the error repair and association relationship construction of model can improve the data processing efficiency by avoiding a large amount of manual operation.In addition, by constructing clothing association relationship network, the data to be fused clothing data is reasonably fused to the corresponding position of system and published, which can help enterprises clearly understand the association between clothing data, and the data of each link is connected from raw material procurement, production process to finished product sales, which can help enterprises make more scientific decisions and improve the smoothness of overall operation management. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the application, and those skilled in the art can also obtain other drawings according to these drawings without creating any creative labor. In the drawings:

[0021] Figure 1A flow chart of a garment data processing method based on an enterprise-level operating system is provided for the embodiments of the present application.

[0022] Figure 2 A structural schematic diagram of a garment data processing device based on an enterprise-level operating system is provided for the embodiments of the present application.

[0023] Reference signs:

[0024] 200: garment data processing device based on an enterprise-level operating system, 201: processor, 202: memory. DETAILED DESCRIPTION

[0025] The embodiments of the present application provide a garment data processing method and device based on an enterprise-level operating system.

[0026] In order for those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0027] The technical solutions proposed by the embodiments of the present application will be described in detail below with reference to the drawings.

[0028] Figure 1 A flow chart of a garment data processing method based on an enterprise-level operating system is provided for the embodiments of the present application, as shown in Figure 1 The garment data processing method based on an enterprise-level operating system includes the following steps:

[0029] S101, input the obtained garment basic data into a preset multi-modal pre-training model to perform entity data recognition and format unification on the garment basic data through the preset multi-modal pre-training model.

[0030] In an implementation manner of the present application, through the preset OCR component, the unstructured data in the garment basic data is extracted for text extraction, and the extracted text and the garment basic data are input into the preset multi-modal pre-training model; wherein the garment basic data includes full-process data of data acquisition, data processing and data storage. Through the preset multi-modal pre-training model, entity data recognition is performed on the garment basic data. Based on the data type corresponding to the entity data, a corresponding preset data format template is determined to perform format unification on the garment basic data through the preset data format template.

[0031] Specifically, the received clothing basic data in the embodiment of the present application includes structured data and unstructured data, and relates to the whole-process data of data acquisition, data processing and data storage. For example, it can include customer data, order data, production data, process data, inventory data, logistics data, scheduling data, financial cost data, supplier procurement data, quality management data, organizational manpower data, master data, etc. It should be noted that the above data is only an example and is not limited thereto. In actual application, the clothing basic data can cover the whole data processing process according to the demand.

[0032] Further, the embodiment of the present application supports data crawling from a specified website or API according to certain specifications to obtain data information. The text information in these unstructured data is recognized and extracted by using the pre-set OCR component, and is converted into a computer understandable text format.

[0033] Further, the extracted text and the original clothing basic data are input into the pre-set multi-modal pre-training model. The pre-set multi-modal pre-training model in the embodiment of the present application is trained by a large amount of data, and can recognize different types of entity data. The entity data can include material entity, process entity, business entity and space-time entity, etc. The material entity refers to various raw materials involved in clothing production, such as cotton, hemp, silk, etc. The process entity covers various processes in the clothing manufacturing process, such as cutting, sewing, printing and dyeing, etc. The business entity includes elements related to the clothing business process, such as order, inventory, sales channel, etc. The space-time entity relates to data related to time and space, such as clothing production cycle, delivery location, etc. It should be noted that the above-mentioned entity data is only an example and is not limited thereto. In actual application, more entity data can be recognized and processed according to the demand.

[0034] Among them, the multi-modal pre-training model in the embodiment of the present application adopts text, image and structured data specific to the clothing industry: such as BOM table, process sheet, for domain adaptation training, and fuses prior knowledge such as clothing terminology library and process rule library. Joint training of text: process description, quality inspection report, and image: design draft, certificate picture, to learn cross-modal semantic alignment, such as: the association between “silk” text and silk fabric image. In the pre-training stage, introduce the clothing industry entity dictionary, such as: fabric type Fabric, process step Step, and relationship rules, such as: process step-uses-equipment, to improve the sensitivity of the model to professional terms.

[0035] Further, based on the identified entity data corresponding to the data type, the corresponding preset data format template is determined. Different types of data have their appropriate standard formats. Utilizing these templates to format the clothing basic data can make the data more consistent and compatible in subsequent processing and storage, facilitating system analysis and management. For example, a built-in regular expression template library is used to automatically identify multi-format data, such as "2023 / 08 / 01" and "01-Aug-2023" unified as "YYYY-MM-DD", solving the problem of industry data heterogeneity.

[0036] In an implementation manner of the present application, based on the preset data detection rule, multi-dimensional data detection is performed on the clothing basic data. The multi-dimensional data detection results are weighted processed to determine the data quality level corresponding to the clothing basic data based on the weighted processing results. Based on different data quality levels, the abnormal data in the clothing basic data is marked in different forms to trigger the corresponding data processing node based on the marking form.

[0037] Specifically, by using the preset data detection rule, the clothing basic data is comprehensively detected from multiple different dimensions. These dimensions cover the accuracy, completeness, consistency, timeliness, etc. of the data. The accuracy detection ensures that the data value is consistent with the actual situation, such as whether the clothing size meets the standard specification; the completeness detection checks whether there is missing data, such as whether some key processes are omitted in the clothing process sheet; the consistency detection ensures the consistency of data in different sources or records, such as whether the color description of the same clothing on different documents is consistent; the timeliness detection judges whether the data is the latest, such as whether the clothing trend data is updated in time, whether the data is expired, such as whether the certificate is expired; traceability: whether the data source is clear, such as whether the design draft is marked with version number and author.

[0038] Further, since the detection results of different dimensions of data have different influences on the overall data quality, the multi-dimensional data detection results are weighted processed. A weight is assigned to the detection result of each dimension, which reflects the relative importance of the dimension in evaluating the data quality. Then the data quality level corresponding to the clothing basic data is determined based on the weighted processing results. Generally, the data quality level can be divided into different levels such as high, medium and low.

[0039] Further, according to different data quality levels, the abnormal data in the clothing basic data is marked in different forms. For example, high quality: completely meets the requirements of the rule library, no missing or conflict, marked as green, directly used in the system. Medium quality: there are slight abnormalities in some fields, but they can be solved by automatic repair. Marked as yellow, used after automatic repair by the system. Low quality: there are serious conflicts or missing of key fields. Marked as red, triggering the manual review process.

[0040] S102, based on the data type of the garment basic data, inputting the garment basic data into different data modules, and performing error data detection on the garment basic data through entity data and knowledge graphs corresponding to different data modules.

[0041] In an implementation form of the present application, based on the data type of the garment basic data, the garment basic data is input into different data modules. For example, the modules of the garment data processing system in the embodiments of the present application can include an AI process recognition module, an AI pattern recognition module, an AI version recognition module, an AI BOM recognition module, and an AI matching rule module, and the like. The garment basic data is classified by type and uploaded to the corresponding module. It should be noted that the above-mentioned data modules are only examples and are not limited thereto. In actual application, more data modules can be added according to functional requirements.

[0042] In an implementation form of the present application, based on the knowledge graph, the logical relationship between different entity data is determined; wherein the logical relationship at least includes one of production relationship, design relationship and quality relationship. Based on the predetermined garment business path, the logical relationship between the garment basic data is detected. The entity data in the garment basic data is compared with the entity relationship in the preset knowledge graph to detect the data logic. Based on different types of entity data, the corresponding historical data is determined, the entity data is compared with the corresponding historical data to determine the data difference. Based on different types of entity data, the corresponding detection error threshold is determined, and based on the detection error threshold and the data difference, the error value detection is performed. Through the business path detection, the data logic detection and the error value detection, the error detection of the garment basic data is realized.

[0043] Specifically, the embodiments of the present application preset a knowledge graph, which represents various entities and their relationships in the form of a graph. In the field of garments, the logical relationship between different entity data can be determined through the knowledge graph. The production relationship reflects the association of various entities in the entire production process from raw material procurement, processing and manufacturing to finished product output; the design relationship involves the connection between elements such as style, color and pattern and other entities in the garment design process; and the quality relationship reflects the relationship between various factors affecting the quality of garments and the final quality of garments. For example, in the knowledge graph, “pure cotton fabric” has a production relationship with “cutting process”, because the cutting process is the first step of processing pure cotton fabric; “fashion style design” has a design relationship with “popular color matching”, which together constitute the design elements of garments.

[0044] Further, the embodiment of the present application is preset with a garment business path, including a standard path of a series of business processes of a garment enterprise from order acquisition, design, procurement, production, quality inspection to sales, etc. By comparing the logical relationship between the garment basic data with the predetermined business path, it is checked whether the business process reflected by the data conforms to the standard path. If the logical relationship between the data does not conform to the predetermined business path, there may be an error. For example, if the historical data is "Supplier A only supplies cotton", the new data "Supplier A-supply-polyester" triggers a relationship path conflict.

[0045] Further, the entity data in the garment basic data is compared with the entity relationship determined in the preset knowledge graph, and it is checked whether the entity relationship in the data conforms to the logic defined by the knowledge graph. If there is a non-conforming case, it means that the data may have a logical error. For example, in the garment production process, the "cutting" process must be prior to the "sewing" process. The start time of the new process data of the sewing process is 09:00, and the cutting process is 10:00, which violates the rule of "cutting prior to sewing".

[0046] For different types of entity data, the corresponding historical data is collected and analyzed. The current entity data is compared with the historical data, and the data difference is calculated. By analyzing the data difference, it can be judged whether the current data has an abnormal wave. For different types of entity data, the corresponding detection error threshold is set, such as adjusting the price fluctuation threshold from 5% to 8%. The detection error threshold refers to the fluctuation range allowed by the data under normal circumstances. The calculated data difference is compared with the detection error threshold. If the data difference exceeds the detection error threshold, it means that the data may have an error.

[0047] Further, through the comprehensive analysis of the business path detection, data logic detection and error value detection, it is comprehensively evaluated whether the garment basic data has an error. If any one of the detection links finds a problem, it can be determined that the data may have an error, which needs to be further reviewed and corrected.

[0048] S103, the repair value of the detected error data is predicted through the data prediction filling model, and the garment data to be fused corresponding to each data module is generated through the repaired data.

[0049] In an implementation manner of the present application, the associated clothing data corresponding to the error data is determined in the clothing basic data, and keyword extraction is performed on the associated clothing data. The text data in the keywords is converted into a low-dimensional dense vector, and the numerical data in the keywords is scaled to a preset numerical section. The converted text data and the scaled numerical data are sorted and combined to construct a model input sequence. The model input sequence is input into a data prediction filling model, and a predicted repair value is obtained through multi-layer Transformer layer operation in the data prediction filling model.

[0050] Specifically, when the error data is identified in the clothing basic data, the associated clothing data needs to be found. These associated data include data of the same batch production, the same style series, or data closely related in the business process. Then, keywords are extracted from the associated clothing data, which can represent the core features of the data, including the style name, size, color, price, production process, and other information of the clothing. For the text data in the extracted keywords, such as clothing style name, brand name, etc., the text data is converted into a vector representation in a low-dimensional vector space through a word embedding algorithm. For the numerical data in the keywords, such as the size and price of the clothing, it is scaled to a preset numerical section, such as scaling the price data to the interval of 0-1, and the size data is mapped to a reasonable numerical range according to the actual situation.

[0051] Further, the converted text data vector and the scaled numerical data are sorted and combined in a certain order to form a sequence as the input of the data prediction filling model. For example, the combination can be performed in the order of brand vector, style vector, color vector, size value, and scaled price value. The constructed model input sequence is input into the data prediction filling model, which usually includes multiple layers of Transformer. Through the operation of the multiple layers of Transformer in the data prediction filling model, the relationships between the elements in the input sequence, such as the association rules between brand, style, color, size, and price, are analyzed, and finally the repair value of the error data is predicted.

[0052] In an implementation manner of the present application, in the case of multiple data source conflicts of error data, the information entropy corresponding to each dimension of the multiple data sources is determined. Based on the information entropy, the weights of the multiple data sources are determined. Based on the weight value, the optimal data source data is determined, and the detected error data is repaired based on the optimal data source data to predict the repair value.

[0053] Specifically, in the clothing data scene, the data provided by different data sources can have differences in multiple dimensions, such as product description, price, inventory quantity, and the like. For each data source, the information entropy of each data dimension is calculated. The more stable and consistent the data of a data source is, the lower the information entropy is; on the contrary, the more chaotic and uncertain the data is, the higher the information entropy is. By calculating the information entropy, the reliability of the data of each data source in each dimension can be quantified.

[0054] After obtaining the information entropy of each data source in different dimensions, a weight is assigned to each data source according to the size of the information entropy, and the lower the information entropy is, the higher the weight of the corresponding data source in the dimension is. The information entropy of each data source in a dimension is normalized, and then 1 is subtracted from the normalized information entropy to obtain the weight of the data source in this dimension. In this way, the weight ranges between 0 and 1, and the sum of the weights of all data sources in the same dimension is 1.

[0055] Further, the weight of each data source in each dimension is weighted and summed with the data value provided by the data source in the dimension to obtain a comprehensive score of each data source, and the data source with the highest score is the optimal data source. The data of the optimal data source is used to predict the repair value of the detected error data in combination with the data prediction filling model.

[0056] S104, the feature vector corresponding to the clothing data to be fused is associated with the feature vector corresponding to the historical clothing data, and a clothing association relationship network is constructed.

[0057] In an implementation manner of the present application, based on the pre-defined data relationship in the knowledge graph, an associated entity graph is constructed, and a plurality of associated entity pairs are determined based on the associated entity graph. The feature vector corresponding to the clothing data to be fused is associated with the feature vector corresponding to the historical clothing data, and based on the association degree, a reference association pair is determined. The reference association pairs existing in the associated entity graph are labeled, and the reference association pairs not existing in the associated entity graph are connected to construct an associated data network. In the historical clothing data, the co-occurrence frequencies corresponding to a plurality of associated data pairs in the associated data network are determined, and the association weights corresponding to the plurality of associated data pairs are determined based on the co-occurrence frequencies. Based on the association degree and the association weight, a comprehensive association degree value corresponding to each associated data is determined, and based on the comprehensive association degree value and the associated data network, a clothing association relationship network is constructed.

[0058] Specifically, the entities in this application embodiment include clothing style, color, size, brand, production process, etc. By analyzing the knowledge graph, multiple related entity pairs are identified. These entity pairs represent two combinations of entities that have a direct or indirect relationship under a predefined relation. Similarity and other correlation indices between the feature vectors of the clothing data to be fused and historical clothing data are calculated to determine reference related pairs. Assuming the clothing data to be fused is a "loose-fitting white T-shirt," its feature vector contains feature information such as fit and color. Historical clothing data contains a large number of records about T-shirts. By calculating the cosine similarity of feature vectors, it is found that "loose-fitting" has a high correlation with "summer casual style," and "white" has a high correlation with "refreshing color scheme." Therefore, "loose-fitting," "summer casual style," "white," and "refreshing color scheme" may become reference related pairs.

[0059] Furthermore, existing reference association pairs in the association entity graph are labeled for differentiation and subsequent analysis. For those pairs that do not exist in the association entity graph but are identified as reference association pairs through association degree analysis, association relationships are established in the association entity graph. By continuously labeling and connecting reference association pairs, a more complete association data network is gradually constructed. In historical clothing data, the co-occurrence frequency of multiple association data pairs in the association data network is statistically analyzed. Co-occurrence frequency refers to the number of times two related entities appear simultaneously in historical data. Association weights are determined based on co-occurrence frequency; the higher the co-occurrence frequency, the greater the association weight. A comprehensive association degree value is obtained by weighted summation of association degree indicators and association weights. Based on these comprehensive association degree values ​​and the association data network, the final clothing association relationship network is constructed.

[0060] In one implementation of this application, based on a function:

[0061]

[0062] Determine the association weights for each of the multiple associated data pairs;

[0063] Among them, w i The association weight is n; the number of associated data pairs is n; the sequence number of the associated data pair is i; f is f. i For co-occurrence frequency; w i α is the association weight; α is the penalty factor; γ is the error. This is a penalty term. The larger the error γ, the more... The smaller the value, the higher the weight w of the corresponding associated data pair. i Reduced; while when the error γ is small, A value close to 1 has little impact on the weights calculated based on co-occurrence frequency.

[0064] S105, based on the clothing association network, the to-be-fused clothing data is fused to the corresponding position of the system, and data publishing is performed.

[0065] In an implementation form of the present application, based on the clothing association network, the corresponding associated historical data of the to-be-fused clothing data is determined. An association mapping is constructed between the to-be-fused clothing data and the associated historical data, and the to-be-fused clothing data is filled into the corresponding position. After the to-be-fused clothing data is filled into the corresponding position, the method further comprises. The to-be-fused data is marked with data information at the corresponding position; wherein the data information at least includes data source and timestamp mark. Based on the marked data information, a relationship iteration chain corresponding to the to-be-fused data is constructed.

[0066] Specifically, by analyzing the key entity information in the to-be-fused clothing data, the historical data with strong association with the to-be-fused clothing data is found in the association network. The corresponding relationship between the to-be-fused clothing data and the associated historical data is established, and after the association mapping is completed, the to-be-fused clothing data is filled into the corresponding position in the system. This position can be a record area related to this type of clothing data in the enterprise resource planning system, the sales database or other data management platform. The process of filling data ensures that the new data can be accurately integrated into the existing data system.

[0067] Further, after the to-be-fused clothing data is filled into the corresponding position, it is marked with data information. The data source mark is used to record from which channel the data is obtained, such as provided by the supplier, collected by market research or generated by the enterprise internal production department. The timestamp mark records the time when the data is generated or updated, which is accurate to the specific date and time. Based on the marked data information, multiple data are arranged and associated in chronological order to obtain the relationship iteration chain corresponding to the to-be-fused data.

[0068] In the embodiment of the present application, the data integration is realized by version management and association expansion without covering the original data:

[0069] 1. Multi-source data merging:

[0070] Scenario: The logistics information of the same order exists in the CSV table (internal system) and the PDF receipt (provided by the supplier) respectively.

[0071] Fusion method: keep two versions of data, establish association relationship (such as ), mark the data source and timestamp.

[0072] 2. History version management:

[0073] Scenario: design draft is modified multiple times (V1.0 to V3.0), and the history version needs to be traced back.

[0074] Fusion mode: store each version in the form of a graph node, and connect them through a "version iteration" relationship chain.

[0075] 3. Cross-modal association expansion:

[0076] Scenario: "embroidery pattern" in image design draft (JPG) and "hand embroidery process" description in process sheet (PDF).

[0077] Fusion mode: extract image features and text keywords, and establish a cross-modal association relationship "design draft-process requirements".

[0078] Non-destructive fusion: retain historical addresses and mark invalid time, add current address, and trace through "historical address" relationship.

[0079] Figure 2 A structural schematic diagram of a garment data processing device based on an enterprise-level operating system provided by an embodiment of the present application is shown in Figure 2 The garment data processing device based on the enterprise-level operating system 200 includes at least one processor 201 and a memory 202 communicatively connected to the at least one processor 201. The memory 202 stores instructions executable by the at least one processor 201. The instructions are executed by the at least one processor 201 to enable the at least one processor 201 to: input the obtained garment basic data into a preset multi-modal pre-training model to perform entity data recognition and format unification on the garment basic data through the preset multi-modal pre-training model; based on the data type of the garment basic data, input the garment basic data into different data modules, and perform error data detection on the garment basic data through the entity data and the knowledge graph corresponding to the different data modules; predict the repair value of the detected error data through a data prediction filling model, generate garment data to be fused corresponding to each data module through the repaired data; perform correlation degree processing on the feature vectors corresponding to the garment data to be fused and the feature vectors corresponding to the historical garment data, and construct a garment correlation relationship network; based on the garment correlation relationship network, fuse the garment data to be fused to the corresponding position of the system, and perform data publishing.

[0080] Each embodiment in the present application is described in a progressive manner, and the same or similar parts of each embodiment can be referred to each other. Each embodiment mainly describes the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0081] The above merely describes the embodiments of the present application, and is not intended to limit the present application. The embodiments of the present application can be variously changed and modified by those skilled in the art. Any modification or replacement does not cause the corresponding technical solution to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A clothing data processing method based on an enterprise-level operating system, characterized by, The method is executed by an enterprise-level operating system, and the method comprises: inputting the obtained clothing basic data into a preset multi-modal pre-training model to perform entity data recognition and format unification on the clothing basic data through the preset multi-modal pre-training model; based on the data type of the clothing basic data, inputting the clothing basic data into different data modules, and performing error data detection on the clothing basic data through the knowledge graph corresponding to the entity data and different data modules, specifically comprising: based on the knowledge graph, determining the logical relationship between different entity data; wherein the logical relationship at least includes one of production relationship, design relationship and quality relationship; based on a predetermined clothing business path, performing business path detection on the logical relationship between the clothing basic data; comparing the entity data in the clothing basic data with the entity relationship in the preset knowledge graph to perform data logic detection; based on different types of entity data, determining corresponding historical data, comparing the entity data with the corresponding historical data to determine a data difference value; based on different types of entity data, respectively determining corresponding detection error thresholds, and based on the detection error thresholds and the data difference value, performing error value detection; through the business path detection, the data logic detection and the error value detection, error detection is realized on the clothing basic data; through a data prediction filling model, predicting a repair value for the detected error data, and generating corresponding to-be-fused clothing data for each data module through the repaired data; performing correlation degree processing on the feature vectors corresponding to the to-be-fused clothing data and the feature vectors corresponding to historical clothing data to construct a clothing correlation relationship network; based on the clothing correlation relationship network, fusing the to-be-fused clothing data to a corresponding position of the system and performing data publishing.

2. The garment data processing method based on an enterprise operating system according to claim 1, characterized in that, The method further comprises: based on a preset data detection rule, performing multi-dimensional data detection on the clothing basic data; ​ ​ ​ 3. The garment data processing method based on an enterprise operating system according to claim 1, characterized in that, ​ ​ The multi-dimensional data detection result is weighted to determine a data quality level corresponding to the clothing basic data based on the weighted result; Based on different data quality levels, different forms of annotation are performed on the abnormal data in the clothing basic data to trigger corresponding data processing nodes based on the annotation form.

4. The garment data processing method based on an enterprise operating system according to claim 1, characterized in that, The error data detected by the data prediction filling model is repaired and predicted, specifically including: Determine the associated clothing data corresponding to the error data in the clothing basic data, and extract keywords from the associated clothing data; Convert the text data in the keywords into a low-dimensional dense vector, and scale the numerical data in the keywords to a pre-set numerical section; Sort and combine the converted text data and the scaled numerical data to construct a model input sequence; Input the model input sequence into the data prediction filling model, and obtain a predicted repair value by operating the multi-layer Transformer layer in the data prediction filling model.

5. The garment data processing method based on an enterprise operating system according to claim 1, characterized in that, The error data detected by the data prediction filling model is repaired and predicted, and the method further includes: In the case of multiple data source conflicts of the error data, determine the information entropy corresponding to the multiple data sources in different dimensions respectively; Based on the information entropy, determine the weight of each data source; Based on the weight value, determine the optimal data source data, and predict the repair value of the detected error data based on the optimal data source data.

6. The garment data processing method based on an enterprise operating system according to claim 1, characterized in that, The feature vector corresponding to the to-be-fused clothing data is associated with the feature vector corresponding to the historical clothing data to construct a clothing association relationship network, specifically including: Based on the pre-defined data relationship in the knowledge graph, an associated entity graph is constructed, and a plurality of associated entity pairs are determined based on the associated entity graph; The feature vector corresponding to the to-be-fused clothing data is associated with the feature vector corresponding to the historical clothing data, and based on the association degree, a reference association pair is determined; The reference association pair existing in the associated entity graph is labeled, and the reference association pair not existing in the associated entity graph is connected to construct an associated data network; In the historical clothing data, the co-occurrence frequency corresponding to each of the plurality of associated data pairs in the associated data network is determined, and the association weight corresponding to each of the plurality of associated data pairs is determined based on the co-occurrence frequency; Based on the association degree and the association weight, a comprehensive association degree value corresponding to each of the plurality of associated data is determined, and the clothing association relationship network is constructed based on the comprehensive association degree value and the associated data network.

7. The garment data processing method based on an enterprise operating system according to claim 6, characterized in that, The association weight corresponding to each of the plurality of associated data pairs is determined based on the function: Determine the association weight corresponding to each of the plurality of associated data pairs; ； Based on the clothing association relationship network, the to-be-fused clothing data is fused into the corresponding position of the system, specifically including: wherein, is an association weight; n is a number of associated data pairs; i is a sequence number of the associated data pair; is a co-occurrence frequency; is an association weight; is a penalty factor; is an error; is a penalty term.

8. The garment data processing method based on an enterprise operating system according to claim 1, characterized in that, Based on the clothing association relationship network, determine the associated historical data corresponding to the to-be-fused clothing data; ​ An association mapping is constructed between the to-be-fused clothing data and the associated historical data, and the to-be-fused clothing data is filled into the corresponding position; After the to-be-fused clothing data is filled into the corresponding position, the method further comprises: Data information of the to-be-fused data is labeled at the corresponding position; wherein the data information at least includes data source and timestamp mark; A relationship iteration chain corresponding to the to-be-fused data is constructed based on the labeled data information.

9. A garment data processing device based on an enterprise-level operating system, characterized by, The device comprises a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to execute the method of any one of claims 1-8.

Images