Intelligent recommendation method based on machine readable standard

Abstract

The invention belongs to the technical field of digitization, and particularly relates to an intelligent recommendation method based on a machine readable standard, and the method comprises the following specific steps: data collection and processing: collecting data related to clothing products and user behaviors from a plurality of channels, removing noise by using a data cleaning technology, correcting errors, and filling missing values; according to the method, a data quality monitoring tool is introduced, an automatic tool is deployed to monitor data integrity and consistency, abnormal data is found and corrected in time, data source diversity is enhanced, multiple reliable data sources are integrated, and the deviation risk caused by a single source is reduced; according to the system, new users can be preliminarily recommended according to demographic information; for a new product, prediction can be made according to the performance condition of similar products, a new possibility is explored while certain precision is ensured, and the system is helped to quickly accumulate effective user behavior data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of digital technology, in particular to an intelligent recommendation method based on machine-readable standards. BACKGROUND

[0002] With the advancement of digital technology, processing and analyzing large-scale clothing data has become more efficient and accurate. The clothing field involves a large amount of product and attribute information, and it is very important for consumers, retailers and supply chain managers to quickly and accurately find the required clothing data. Machine-readable standards can help build a clothing knowledge graph, so that data is stored and represented in a structured manner, providing more efficient data search and access capabilities. Through standardized data representation and semantic association, technical applications such as interoperability, intelligent recommendation and personalized customization of clothing data can be achieved, thereby promoting the digital transformation, intelligent development and user experience improvement of the clothing field, bringing more benefits and business opportunities to enterprises;

[0003] The intelligent recommendation method based on machine-readable standards refers to using structured, standardized data formats and advanced algorithms to analyze user behavior patterns, preferences and content features, thereby providing personalized recommendations for users. This method relies on well-defined data models and metadata standards, allowing computers to understand and process information, thereby enabling more accurate content recommendations;

[0004] However, the existing recommendation methods still have the following technical problems when used:

[0005] Data quality is uneven, and incorrect or incomplete data can affect recommendation accuracy. For new users or new projects, it is difficult to make accurate recommendations due to the lack of sufficient historical data.

[0006] Therefore, the present application proposes an intelligent recommendation method based on machine-readable standards to solve the above-mentioned problems. SUMMARY

[0007] The purpose of the present application is to provide an intelligent recommendation method based on machine-readable standards to solve the problems mentioned in the background.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical solution: an intelligent recommendation method based on machine-readable standards, the specific steps of which are as follows:

[0009] Step 1, data collection and processing: collect data related to clothing products and user behavior from multiple channels, use data cleaning techniques to remove noise, correct errors, and fill in missing values, and convert all raw data into a unified machine-readable format to ensure consistency in subsequent processing;

[0010] Step two, feature extraction: extract key product attributes from clothing product information, analyze user's historical purchase records, browsing behavior, favorites information, build user portrait, identify user's preference pattern, understand the influence of time, weather, geographical location factors on user selection, increase context-related feature dimensions;

[0011] Step three, product recommendation: for new users or new products, make preliminary recommendations by analyzing their description information, use the behavior of similar interest groups in the user's friend circle or attention list as a reference, provide personalized suggestions for new users, combine the advantages of collaborative filtering, content-based recommendation and algorithm to reduce the limitations of single algorithm in cold start scenario;

[0012] Step four, model training and optimization: select appropriate machine learning or deep learning algorithms according to actual needs, use cross-validation techniques to evaluate model performance, and optimize model parameters through grid search or random search, design online learning framework to update model weights with new data and adapt to dynamic market environment;

[0013] Step five, recommendation generation and display: based on user portrait and current context information, real-time generation of personalized clothing recommendation list, using a variety of display methods;

[0014] Step six, user feedback: encourage users to provide direct feedback and indirectly understand user satisfaction through implicit feedback, regularly analyze feedback data, adjust recommendation strategy, optimize algorithm performance, and form a virtuous cycle.

[0015] Preferably, in step one, the data collection channels include e-commerce platforms, social media and user feedback;

[0016] When processing data, specifically using statistical methods or machine learning models to predict and complete missing data points;

[0017] And the readable format of the original data conversion is XML, JSON, RDF.

[0018] Preferably, in step two, extracting key attributes from clothing product information includes style, color, material and brand.

[0019] Preferably, the description information in step three includes text description and image features.

[0020] Preferably, in step four, image features are extracted by matrix decomposition, neural collaborative filtering and convolutional neural network.

[0021] Preferably, the display method in step five includes sorting by style, sorting by price and recommending by popular trends.

[0022] Preferably, in step six, feedback can be directly provided through customer ratings and reviews, and indirect understanding of user satisfaction includes click-through rate and dwell time.

[0023] Compared with the prior art, the present application has the following advantages:

[0024] 1) The present method introduces data quality monitoring tools, deploys automated tools to monitor data integrity and consistency, discovers and corrects abnormal data in a timely manner, enhances data source diversity, integrates multiple reliable data sources, and reduces the bias risk caused by a single source.

[0025] 2) For new users, the present system can make preliminary recommendations based on demographic information; for new products, it can make predictions based on the performance of similar products, explore new possibilities while ensuring a certain level of accuracy, and help the system quickly accumulate effective user behavior data. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 The flowchart of the present method is shown in the accompanying drawings. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0028] Embodiment:

[0029] Please refer to Figure 1 The present application provides a technical solution:

[0030] A kind of intelligent recommendation method based on machine readable standard, the specific step flow of the intelligent recommendation method based on machine readable standard is as follows:

[0031] Step one, data collection and processing: by collecting data related to clothing products and user behavior from multiple channels, using data cleaning techniques to remove noise, correct errors, and fill in missing values, converting all raw data into a unified machine-readable format, ensuring consistency in subsequent processing;

[0032] The data collection and processing stage is the foundation of the entire recommendation system, and its quality directly affects the accuracy, personalization and system stability of the recommendation. In the clothing recommendation scenario, the data sources are diverse, the structure is complex, and there are many noise disturbances.

[0033] The collected channels include e-commerce platforms, social media, user feedback, IoT wearable devices, and ERP systems, but regardless of the collection method, it needs to comply with relevant laws and regulations to ensure data collection methods and user authorization;

[0034] When collecting data, it is necessary to ensure that the collected data covers different user groups (age, gender, region, etc.) and different clothing categories (men's clothing, women's clothing, sports, business attire, etc.), and prepare enough auxiliary data for new users and new goods;

[0035] Some clothing recommendations need to consider dynamic factors such as season, fashion trends, and promotional activities, and need to update data regularly, and retain a certain period of historical data for trend analysis and model training;

[0036] When processing data, use statistical methods or machine learning algorithms to identify abnormal data, delete abnormal data, correct data, and mark abnormal data for manual review; identify missing fields and handle them by using mean, median, mode to fill; using the most common value or "Unknown" label to fill; using NLP or image generation technology for synthetic filling;

[0037] In the collected data set, according to the unique identifier, remove duplicates, when user behavior data from different platforms conflict, use timestamp, credibility weighting method to merge.

[0038] During data processing, sensitive user information (such as mobile phone number, address) is desensitized or encrypted, access permissions are set for data during processing to prevent data leakage, and non-essential user identity information is anonymized.

[0039] Step two, feature extraction: extract key product attributes from clothing product information, analyze user's historical purchase records, browsing behavior, and favorites information, build user portraits, identify user preference patterns, understand the influence of time, weather, and geographical location factors on user selection, and increase context-related feature dimensions;

[0040] Feature extraction is one of the core steps in building a recommendation system, responsible for extracting key features from clothing product information and user behavior data that help the recommendation model understand user preferences and product characteristics, and combining context information (such as time, weather, and geographical location) to enhance the personalization and scenario adaptation capabilities of recommendations;

[0041] In structured attribute extraction, the extraction dimensions include style, color, material, brand, price range, and suitable occasions. The standardized representation uses a unified label system, such as "style" unified as "simple", "street", "retro" and other standard labels. In image feature extraction, CNN is used to extract visual features of clothing images for visual similarity recommendation. In text feature extraction, NLP technology is used to extract semantic features from product descriptions, reviews, and label text information.

[0042] In user feature extraction, the following points should be noted:

[0043] 1. Behavior data modeling

[0044] Explicit feedback: purchase records, ratings, reviews, and other direct reflections of user preferences.

[0045] Implicit feedback: browsing duration, click frequency, favorites, shopping cart behavior, and other indirect preference signals.

[0046] 2. User portrait construction

[0047] Static attributes: age, gender, region, occupation, and other basic information.

[0048] Dynamic interest labels: generate "preference labels" based on user behavior, such as "like Korean fashion" and "prefer high-waisted jeans".

[0049] Time series modeling: use LSTM, Transformer and other models to capture the trend of user interest over time.

[0050] 3. Handling of cold start users

[0051] Initial modeling based on registration information: use static information such as age, gender, and region to construct initial interest labels.

[0052] Social graph assisted modeling: use the behavior data of similar users in the social network to make inferences.

[0053] Step three, product recommendation: for new users or new products, analyze their description information to make preliminary recommendations, use the behavior of similar interest groups in the user's friend circle or follow list as a reference, and provide personalized suggestions for new users, combining the advantages of collaborative filtering, content-based recommendation and algorithm to alleviate the limitations of a single algorithm in the cold start scenario.

[0054] Recommendations for new users:

[0055] Ensure product description structure, standardization: new user no behavior data, rely on machine-readable metadata (such as style, color, style label) of goods for matching; Must ensure that these attributes have been semantically unified through pre-processing (such as Korean style ≠ Korean version ≠ Korean trend);

[0056] Use lightweight matching model:

[0057] Cosine similarity, Jaccard similarity and other fast calculation of new user registration information (such as gender, age, region) and the matching degree of product label can be used;

[0058] Avoid over-reliance on a single attribute:

[0059] For example, only according to "female" to recommend all women's clothing, should be combined with age, occupation and other factors to build an initial interest portrait.

[0060] Recommendations for new products:

[0061] High-quality feature extraction is a prerequisite: new products have no interaction data, must rely on their text description, image, label and other static information, need to ensure that the pre-feature extraction module has generated accurate vector representation for them;

[0062] Semantic matching with existing products: associate new products with similar products in the database (such as "same style, same material, same application scenario"), inherit their recommendation path, example: a "white cotton V-neck T-shirt" can be matched to the recommendation pool of the historical best-selling "pure white basic round-neck T-shirt".

[0063] Active exposure and exploration mechanism:

[0064] Assign a certain exposure opportunity to new products in the recommendation list to collect initial user feedback; Balance the ratio of "using known preferences" and "exploring unknown products".

[0065] Set up a cold start support strategy: give new products a short-term traffic tilt (such as front page recommendation position, new person exclusive area), accelerate the accumulation of behavior data.

[0066] Mixed recommendation strategy fusion considerations:

[0067] 1. Fusion of collaborative filtering (CF) and content-based recommendation

[0068] Avoid the failure of collaborative filtering in cold start: collaborative filtering relies on user-item interaction matrix, new users / new goods are "sparse points" in this matrix, direct use is poor;

[0069] Use hybrid model architecture: initially give priority to "content-based" (weight 70%), gradually increase the weight of collaborative filtering as data accumulates;

[0070] Use Wide & Deep, Two-Tower model to learn content features and collaborative signals simultaneously.

[0071] Example: User Tower inputs user attributes + social preferences, Item Tower inputs item semantic features, jointly train matching scores.

[0072] 2. Algorithm selection and explainability

[0073] Avoid "black box" decisions: cold start recommendations should have some explainability, such as: "We recommend this dress for you because the temperature in your city today is 28°C, suitable for summer clothes."

[0074] Support rule engine for backup:

[0075] When the model confidence is low, enable rule-based recommendations (such as summer → recommend short sleeves) as a supplement.

[0076] Step four, model training optimization: according to the actual demand, select appropriate machine learning or deep learning algorithm, use cross-validation technology to evaluate model performance, and optimize model parameters through grid search or random search, design online learning framework, make model update its weight with new data, adapt to dynamic changing market environment;

[0077] Step five, recommendation generation and display: based on user portrait and current context information, real-time generation of personalized clothing recommendation list, using multiple display methods;

[0078] Step six, user feedback: encourage users to provide direct feedback, and indirectly understand user satisfaction through implicit feedback, regularly analyze feedback data, adjust recommendation strategy, optimize algorithm performance, form a virtuous cycle.

[0079] In step one, the data collection channels include e-commerce platforms, social media and user feedback;

[0080] When processing data, specifically using statistical methods or machine learning models to predict and complete missing data points;

[0081] And the readable format of the original data conversion is XML, JSON, RDF.

[0082] In step two, the key attributes extracted from clothing product information include style, color, material and brand.

[0083] The description information in step three includes text description and picture features.

[0084] In step four, image features are extracted through matrix decomposition, neural collaborative filtering and convolutional neural networks.

[0085] The display mode in the fifth step includes classification by style, sorting by price, and recommendation by popular trend.

[0086] In the sixth step, feedback can be provided directly through ratings and reviews provided by customers, and indirectly through click rate and dwell time to understand user satisfaction.

[0087] In the description of the present application, it should be understood that the indicated orientation or positional relationship is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0088] The basic principles and main features of the present application and the advantages of the present application are shown and described above, and it is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application; therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application, and any reference signs in the claims should not be regarded as limiting the claims involved.

[0089] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An intelligent recommendation method based on machine-readable standards, characterized in that, The specific steps of this intelligent recommendation method based on machine-readable standards are as follows: Step 1: Data Collection and Processing: By collecting data on apparel products and user behavior from multiple channels, data cleaning techniques are used to remove noise, correct errors, and fill in missing values. All raw data is converted into a unified machine-readable format to ensure consistency in subsequent processing. Step 2, Feature Extraction: Extract key product attributes from clothing product information, analyze users' historical purchase records, browsing behavior, and favorites information to build user profiles, identify user preference patterns, understand the impact of time, weather, and geographical location factors on user choices, and add context-related feature dimensions. Step 3, Product Recommendation: For new users or new products, make initial recommendations by analyzing their description information. Use the behavior of similar interest groups in the user's social circle or following list as a reference to provide personalized suggestions for new users. Combine the advantages of collaborative filtering, content-based recommendation and algorithms to reduce the limitations of a single algorithm in cold start scenarios. Step 4: Model Training and Optimization: Select appropriate machine learning or deep learning algorithms according to actual needs, use cross-validation to evaluate model performance, optimize model parameters through grid search or random search, and design an online learning framework so that the model can continuously update its weights as new data arrives to adapt to the dynamically changing market environment. Step 5: Recommendation Generation and Display: Based on user profiles and current context information, a personalized clothing recommendation list is generated in real time, using multiple display methods; Step Six: User Feedback: Encourage users to provide direct feedback and indirectly understand user satisfaction through implicit feedback. Regularly analyze feedback data, adjust recommendation strategies, optimize algorithm performance, and form a virtuous cycle.

2. The intelligent recommendation method based on machine-readable standards according to claim 1, characterized in that: In step one, the data collection channels include e-commerce platforms, social media, and user feedback; In data processing, statistical methods or machine learning models are used to predict and fill in missing data points. Furthermore, the original data is converted to readable formats such as XML, JSON, and RDF.

3. The intelligent recommendation method based on machine-readable standards according to claim 1, characterized in that: In step two, key attributes, including style, color, material, and brand, are extracted from the clothing product information.

4. The intelligent recommendation method based on machine-readable standards according to claim 1, characterized in that: The descriptive information in step three includes text descriptions and image features.

5. The intelligent recommendation method based on machine-readable standards according to claim 1, characterized in that: In step four, image features are extracted through matrix factorization, neural collaborative filtering, and convolutional neural networks.

6. The intelligent recommendation method based on machine-readable standards according to claim 1, characterized in that: The display methods in step five include categorizing by style, sorting by price, and recommending by trend.

7. The intelligent recommendation method based on machine-readable standards according to claim 1, characterized in that: In step six, feedback can be directly provided through customer ratings and comments, while indirect ways to understand user satisfaction include click-through rate and dwell time.

Images