CN117911114A

CN117911114A - Clothing fashion trend capturing and recommending method based on big data

Info

Publication number: CN117911114A
Application number: CN202410092196.5A
Authority: CN
Inventors: 赵冠华; 冯冉冉; 马伊辰; 朱可欣
Original assignee: Inner Mongolia University of Technology
Current assignee: Inner Mongolia University of Technology
Priority date: 2024-01-22
Filing date: 2024-01-22
Publication date: 2024-04-19

Abstract

The invention discloses a clothing fashion trend capturing and recommending method based on big data. The method acquires and analyzes text data related to clothing in real time from unstructured text of different data sources. These data include clothing sales volume data, user rating data, price change data, and user interaction data. By analyzing this data, the method is able to extract keywords on fashion trends of clothing and further analyze these keywords to identify the trend of the fashion elements. Based on these trend analyses, the method can predict future fashion trends of clothing. In addition, the method generates a personalized clothing recommendation list based on historical purchase data and browsing behaviors of the user and in combination with predicted popularity trends. The method provided by the invention provides an innovative way for the clothing industry to capture and respond to rapidly changing fashion trends, and simultaneously provides more personalized shopping experience for consumers.

Description

Clothing fashion trend capturing and recommending method based on big data

Technical Field

The invention relates to the technical field of data processing, in particular to a clothing fashion trend capturing and recommending method based on big data.

Background

In the current clothing industry, it is critical to quickly capture and respond to popular trends. With the advent of the internet and social media, vast amounts of data were generated that contained important information about consumer preferences, market trends, and fashion dynamics. However, since such data is typically unstructured, such as user comments, social media posts, sales records, etc., it is a challenge to extract valuable information and identify popular trends therefrom. In addition, rapid changes in the apparel industry require retailers and designers to not only learn current trends in popularity, but also predict future trends in order to more effectively plan and respond to market changes.

Conventional trend analysis methods often rely on expert subjective judgment and historical sales data, which may result in insufficiently timely or lack of personalization of the response to emerging trends. With the development of artificial intelligence and big data analysis techniques, new opportunities have emerged to address these issues. In particular, natural Language Processing (NLP) technology has become a powerful tool for processing and analyzing large amounts of text data, which can be used to automatically extract and analyze information related to fashion trends of clothing.

Therefore, it is highly desirable to develop a method for capturing and recommending fashion trends of clothing based on big data.

Disclosure of Invention

The application provides a clothing fashion trend capturing and recommending method based on big data, which is used for improving clothing recommending accuracy.

The application provides a clothing fashion trend capturing and recommending method based on big data, which comprises the following steps:

Acquiring clothing related text data from unstructured texts of different data sources in real time, wherein the clothing related text data comprise clothing sales volume data, user evaluation data, price change data and user interaction data;

Analyzing the clothing-related text data using natural language processing techniques to extract keywords related to fashion trends of clothing;

Analyzing the keywords related to the fashion trends of the clothing to identify the rising or falling trend of the fashion elements, and predicting the fashion trends of the clothing in the future according to the identified rising or falling trend of the fashion elements;

Based on the historical purchase data and browsing behaviors of the user and the predicted fashion trends of the clothing, a personalized clothing recommendation list is generated.

Still further, the analyzing the clothing-related text data using natural language processing techniques to extract keywords regarding clothing popularity trends includes:

Inputting the clothing related text data into a trained dynamic semantic analysis model to obtain keywords of clothing popular trends;

the dynamic semantic analysis model comprises a preprocessing module, a local feature extraction module, a sequence feature extraction module, a keyword weight adjustment mechanism, a feature selection and optimization mechanism and a data fusion module;

The preprocessing module is used for receiving the user evaluation data and the user interaction data, and carrying out standardized processing on the received data to obtain preprocessed text data;

The local feature extraction module is used for receiving the preprocessed text data, extracting local features in the preprocessed text data by using a convolutional neural network, and obtaining high-level representation of the local features;

The sequence feature extraction module is used for receiving the advanced representation of the local feature, capturing long-distance dependence and dynamic change in the received data by using a cyclic neural network, and obtaining a dynamic feature representation, wherein the dynamic feature representation comprises context information of keywords;

the keyword weight adjustment mechanism is used for receiving the dynamic feature representation, performing emotion analysis on each vocabulary in the dynamic feature representation by using a pre-trained deep learning emotion analysis model, and adjusting the weight of each vocabulary in the final keyword extraction according to emotion tendencies to obtain emotion weighted text features;

The feature selection and optimization mechanism is used for receiving emotion weighted text features, optimizing the emotion weighted text features by using a gradient-based feature selection method, and obtaining optimized text features;

The data fusion module is used for receiving the optimized text characteristics, the clothing sales volume data and the price change data, and fusing and analyzing the received data to obtain keywords about clothing fashion trends.

Further, the analyzing the keywords related to the fashion trend of the clothing to identify the rising or falling trend of the fashion element, and predicting the fashion trend of the clothing in the future according to the identified rising or falling trend of the fashion element, includes:

Calculating a trend score of the keywords based on the occurrence frequency and the growth rate of the keywords with respect to the fashion trend of the clothing;

Clustering the keywords according to the trend scores of the keywords to form different popular trend categories, and selecting representative keywords from each category;

future popularity trends are predicted by time series analysis using trend scores of selected representative keywords.

Still further, the calculating the trend score of the keyword based on the occurrence frequency and the growth rate of the keyword with respect to the fashion trend of the clothing includes:

the trend Score (k) of the keyword k is calculated using the following equation 1:

Wherein F _k is the occurrence frequency of the keyword k, and F _k is obtained by dividing the number of times the keyword appears in a specified time window by the total number of words; SS _k is the emotion score of keyword k, and the calculation method of SS _k is to subtract the number of negative comments from the number of positive comments referring to the keyword, and divide the obtained difference by the total number of comments; alpha, beta and gamma are weight parameters; GR _k is the growth rate of the keyword k, and GR _k is calculated using the following formula 2:

Wherein F _k,current is the occurrence frequency of the keyword k in the current time window; f _k,previous is the frequency of occurrence of the keyword k in the last time window.

Still further, the generating a personalized clothing recommendation list based on the historical purchase data and browsing behavior of the user and the predicted clothing popularity trend includes:

collecting historical purchase data and browsing behaviors of a user, wherein the historical purchase data comprises clothing types and brands purchased by the user in the past; the browsing behavior comprises browsing history of a user on an electronic commerce platform;

Identifying user preferences for specific clothing styles, styles and categories according to the user's historical purchase data and browsing behavior;

Combining the identified preferences for specific clothing styles, styles and categories with predicted fashion trends to generate a list of clothing recommendations for the user.

Still further, the collecting historical purchase data and browsing behavior of the user includes:

automatically tracking and recording browsing behaviors and purchasing activities of a user by using login information and browser data of the user on an electronic commerce platform;

The collected data is stored in the user profile for subsequent data analysis and garment recommendation list generation.

Still further, the identifying user preferences for specific clothing styles, styles and categories based on the user's historical purchase data and browsing behavior includes:

Analyzing the stay time, the click frequency and the search history of the user browsing the page to identify the interest intensity of the user on different products;

the user purchase history is analyzed to determine the brand, style, and price range of the user's preferences.

Still further, the combining the identified preferences for specific clothing styles, styles and categories with predicted fashion trends for clothing generates a list of clothing recommendations for the user, including:

and generating a personalized recommendation list by using a collaborative filtering recommendation algorithm according to the user preference and the predicted fashion trend of the clothing.

The application provides a characteristic generating device of network flow data, comprising:

A processor;

And a memory for storing a program which, when read by the processor, performs the big data based clothing popularity trend capturing and recommending method as described above.

The present application provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs the method of fashion trend capture and recommendation for clothing based on big data as described above.

The technical scheme provided by the application has the beneficial effects that:

(1) By acquiring and analyzing unstructured text data from different data sources in real time, such as clothing sales, user ratings, price changes, and user interaction data, the method can provide comprehensive market insight. This comprehensiveness enables apparel brands and retailers to quickly adapt to market changes, better understanding consumer needs and preferences. (2) Text data is analyzed using natural language processing techniques to extract keywords related to fashion trends of clothing and trend the keywords, the method being able to accurately identify current and future popular elements. This accurate trend prediction is very valuable to clothing designers, retailers, and market analyzers because it helps them know which elements may be the focus of attention for the consumer. (3) By combining historical purchase data and browsing behavior of the user with predicted fashion trends of the clothing, the method is able to generate highly personalized clothing recommendation lists. This not only enhances the shopping experience of the user, but also helps to increase consumer satisfaction and loyalty. (4) Providing data-based insight to apparel brands and retailers, helping them make more intelligent business decisions such as product development, inventory management, and market positioning.

Drawings

Fig. 1 is a flowchart of a method for capturing and recommending fashion trends of clothing based on big data according to a first embodiment of the present application.

Fig. 2 is a schematic diagram of a dynamic semantic analysis model according to a first embodiment of the present application.

Detailed Description

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application may be embodied in many other forms than those herein described, and those skilled in the art will readily appreciate that the present application may be similarly embodied without departing from the spirit or essential characteristics thereof, and therefore the present application is not limited to the specific embodiments disclosed below.

The first embodiment of the application provides a clothing fashion trend capturing and recommending method based on big data. Referring to fig. 1, a schematic diagram of a first embodiment of the present application is shown. The following provides a detailed description of a fashion trend capturing and recommending method based on big data according to a first embodiment of the present application with reference to fig. 1.

Step S101: and acquiring clothing related text data from unstructured texts of different data sources in real time, wherein the clothing related text data comprise clothing sales volume data, user evaluation data, price change data and user interaction data.

In step S101, a method of acquiring unstructured text data associated with a garment in real time from a plurality of data sources is implemented. This step is the basis of this embodiment, as it provides raw data for subsequent data analysis and trend prediction. The detailed description is as follows:

This step involves identifying and selecting a data source. Data sources mainly include, but are not limited to, electronic commerce platforms, social media platforms, fashion blogs, forums, and the like. These platforms are chosen because they are the primary channels for capturing current clothing trends and consumer preferences. Each data source should be sufficiently representative and data rich to ensure that the information collected is fully reflective of market trends.

In this step, the referred garment related text data includes, but is not limited to, garment sales data, user rating data, price change data, and user interaction data. These data types are chosen because they provide direct information about consumer behavior and market reactions. In particular, garment sales data may reflect popularity of a style, user ratings data typically includes feedback on style, quality, and overall satisfaction, price change data may reveal market demand and inventory levels, and user interaction data (e.g., praise, comment, and share) provides insight about consumer interests and community revenues.

The step adopts data grabbing technology and API (application programming interface) call to realize the acquisition of real-time data. This involves writing specialized scripts or using off-the-shelf data crawling tools that can automatically access a designated website or platform, retrieve updated content, and crawl relevant text data. To ensure real-time and accuracy of the data, the system performs data grabbing operations periodically (e.g., every hour or every day) to capture the latest market dynamics.

The step also includes preprocessing the raw data to facilitate subsequent analysis, considering that the collected data is unstructured. Preprocessing operations may include data cleansing (removal of extraneous content, duplicate items, etc.), formatting (unifying data formats, facilitating storage and processing), and preliminary classification (e.g., marking by data type, source).

Step S101 provides the necessary input data for the subsequent steps S102 to S104. By effectively collecting and preprocessing a large amount of unstructured text data, a foundation is laid for deep analysis, epidemic trend identification and personalized recommendation generation by using natural language processing technology.

In summary, step S101 is a critical loop in this embodiment, which provides a solid basis for capturing and analyzing fashion trends of clothing by accurately and efficiently collecting and preprocessing a large amount of relevant data.

Step S102: the garment related text data is analyzed using natural language processing techniques to extract keywords regarding garment popularity trends.

Step S102 is a key element in the embodiment, and involves analyzing text data using Natural Language Processing (NLP) technology to extract keywords related to fashion trends of clothing.

In this step, unstructured text data collected in step S101 is processed using natural language processing techniques. Natural language processing techniques include, but are not limited to, text mining, emotion analysis, keyword extraction, and semantic analysis. These techniques enable the system to understand and analyze complex patterns in human language, thereby effectively extracting information about fashion trends in clothing.

First, the collected text is preprocessed, including word normalization, stop word removal, part-of-speech tagging, etc., to prepare the data for further analysis. Keyword extraction algorithms (e.g., TF-IDF, LDA) are then used to identify keywords in the text. These keywords are key to understanding and analyzing fashion trends in garments, and may include specific garment styles, popular colors, fabric types, or design elements.

Emotional analysis is performed on user rating data and comments to assess the public's attitudes (positive, negative, or neutral) to a certain garment style. And meanwhile, the context information in the text is understood by utilizing a semantic analysis technology, so that the true meaning of the keyword can be accurately identified, and ambiguity is avoided.

This step can be implemented by algorithms and computational models that are optimized to process large amounts of text data and extract the most valuable information therefrom.

Step S102 is a key step of converting unstructured text data into meaningful, operational information. By accurately extracting and analyzing the keywords, a necessary basis is provided for trend recognition and prediction in step S103. In addition, the extracted information directly affects the accuracy and relevance of the personalized recommendation list in step S104.

In this embodiment, the analyzing the clothing-related text data using natural language processing technology to extract keywords about fashion trends of clothing includes:

Inputting the clothing related text data into a trained dynamic semantic analysis model 200 to obtain keywords of clothing fashion trends; the dynamic semantic analysis model comprises a preprocessing module 201, a local feature extraction module 202, a sequence feature extraction module 203, a keyword weight adjustment mechanism 204, a feature selection and optimization mechanism 205 and a data fusion module 206;

The preprocessing module 201 is configured to receive the user evaluation data and the user interaction data, and perform standardization processing on the received data to obtain preprocessed text data; the local feature extraction module 202 is configured to receive the preprocessed text data, and extract local features in the preprocessed text data by using a convolutional neural network, so as to obtain a high-level representation of the local features; the sequence feature extraction module 203 is configured to receive the advanced representation of the local feature, and capture long-distance dependence and dynamic variation in the received data by using a recurrent neural network, to obtain a dynamic feature representation, where the dynamic feature representation includes context information of a keyword; the keyword weight adjustment mechanism 204 is configured to receive the dynamic feature representation, perform emotion analysis on each vocabulary in the dynamic feature representation by using a pre-trained deep learning emotion analysis model, and adjust the weight of the vocabulary in the final keyword extraction according to emotion tendencies to obtain emotion weighted text features; the feature selection and optimization mechanism 205 is configured to receive emotion weighted text features, and optimize the emotion weighted text features by using a gradient-based feature selection method to obtain optimized text features; the data fusion module 206 is configured to receive the optimized text feature, clothing sales data, and price change data, and fuse and analyze the received data to obtain keywords related to fashion trends of clothing.

In this embodiment, one of the core components of the dynamic semantic analysis model 200 is a preprocessing module 201, which plays a vital role. This module is specifically responsible for processing and preparing the garment related text data into the model, ensuring that the data quality and format are suitable for subsequent deep analysis.

The preprocessing module 201 first receives raw text data from a variety of sources including, but not limited to, user ratings and user interaction data. User ratings typically relate to mindsets, favorites and discontents on clothing, while user interaction data may include textual content of actions like praise, comment or share. These data are in a variety of forms and may contain various non-standardized elements such as slang, abbreviations or erroneous grammatical structures, and therefore require extensive pre-processing for subsequent analysis.

The first step of the preprocessing module 201 is data cleansing. This step includes removing extraneous information in the text, such as advertisements, meaningless filler words or special symbols, and the like. In addition, text normalization is also performed at this stage, including unifying vocabulary cases, correcting spelling errors, and converting network slang or abbreviations to their standard form. The text processed in this way is cleaner and more standard, and is convenient for subsequent machine processing.

The preprocessing module 201 then performs a word embedding process on the text, i.e., converting the text into a machine-understandable form. This typically involves converting each word into a vector of a certain dimension. These vectors represent the semantic features of the vocabulary and can be efficiently processed by the deep learning model. Word embedding not only improves the expressive power of text data, but also helps capture complex relationships between words.

Finally, the preprocessing module 201 passes these cleaned, normalized and embedded text data to the next stage of the model, namely the local feature extraction module 202. The preprocessed text data has the condition of being further analyzed by the deep learning model, and a solid foundation is laid for extracting keywords of fashion trends of clothing.

The preprocessing module 201 is a vital loop in the dynamic semantic analysis model. Through thorough cleaning, standardized processing and word embedding conversion of the original text data, accurate and high-quality input data is provided for the whole model, and the effectiveness and accuracy of subsequent analysis are ensured.

In this embodiment, one key component of the dynamic semantic analysis model 200 is the local feature extraction module 202. The main responsibility of this module is to extract valuable local features from the pre-processed text data, which lays a solid foundation for further depth analysis. The details of the implementation of the local feature extraction module 202 are as follows:

once the preprocessing module 201 completes the normalization process of the garment related text data, the data is passed to the local feature extraction module 202. The core of this module is the use of Convolutional Neural Networks (CNNs) to process text data. Convolutional neural networks are particularly useful for identifying and extracting local patterns and features in data, making them ideal choices for processing text data.

In the process, the CNN first operates on the input text data through a series of convolution layers. Each convolution layer contains a plurality of convolution kernels that slide over the text data and extract key local features, such as specific lexical combinations or grammatical structures. This process can be analogous to looking up meaningful patterns and keywords in text. The pooling layer is then used to reduce the dimensionality of these features, thereby improving data processing efficiency and reducing the consumption of computing resources.

After the pooling layer processing, the extracted local features are converted into high-level representations that capture important information in the text, while rejecting redundant and irrelevant details. This high-level representation contains not only important local features of the text data, but also retains a sufficient amount of information for further analysis by the sequence feature extraction module 203.

The design of the local feature extraction module 202 enables the model to efficiently identify and extract key information in text that is critical to understanding clothing-related discussions and comments. By accurately extracting these local features, the module provides strong support for capturing subtle trends and patterns in the text data.

The local feature extraction module 202 effectively extracts key local features from the text data through advanced convolutional neural network techniques, which are then used to in-depth analyze and understand clothing popularity trends. The implementation of the module is an indispensable part of a dynamic semantic analysis model, and provides a solid foundation for subsequent text analysis.

In the present embodiment, another key element of the dynamic semantic analysis model 200 is the sequence feature extraction module 203, which is responsible for in-depth analysis of the advanced feature representations obtained from the local feature extraction module 202. The implementation details of the module are as follows:

The main task of the sequence feature extraction module 203 is to process and analyze local features extracted by Convolutional Neural Networks (CNNs) to capture long-term dependencies and dynamic changes in text data. This process is critical because it enables the model to understand the contextual information in the text, thereby more accurately identifying and extracting keywords about fashion trends of clothing.

To achieve this goal, the sequence feature extraction module employs a Recurrent Neural Network (RNN). RNNs are a type of neural network specifically designed for processing sequence data that delivers previous information through cyclic connections between nodes, thereby preserving context information at different points in time of the sequence. In processing text data, RNNs enable models to understand the overall meaning and structure of sentences by considering the order and interrelationships of words in sentences.

In an implementation, the RNN receives as input a local feature high-level representation from the CNN. These features are first fed into each of the cyclic units of the RNN. In each cell, the RNN combines the currently entered features and hidden states from the previous cell to generate a new hidden state, which is repeated throughout the text sequence. In this way, the RNN can comprehensively consider information in the entire text sequence, capturing dynamic changes of keywords over time and contextual relationships.

Finally, the sequence feature extraction module outputs a dynamic feature representation that contains contextual information of the keywords in the text data. These dynamic feature representations provide key information for subsequent analysis, including emotion analysis and feature selection.

The sequence feature extraction module 203 is a vital part of the dynamic semantic analysis model. It exploits the powerful capabilities of recurrent neural networks to process and analyze local feature representations to capture contextual information and long-term dependencies in text data. In this way, the module provides the necessary basis for a deeper understanding and analysis of fashion trends in garments.

In this embodiment, the keyword weight adjustment mechanism 204 is an important component of the dynamic semantic analysis model, and its main function is to enhance the judgment capability of the model on the importance of keywords in text data through emotion analysis. Specifically, a detailed description of how this mechanism is implemented is as follows:

When the sequence feature extraction module 203 completes capturing the long-distance dependency and dynamic change in the text data, a dynamic feature representation containing the keyword context information is obtained. These dynamic feature representations are then passed to the keyword weight adjustment mechanism 204. At this stage, the task of the model is to conduct in-depth emotion analysis on these dynamic feature representations to further refine and optimize the keyword extraction process.

To achieve emotion analysis, keyword weight adjustment mechanism 204 integrates a pre-trained deep learning emotion analysis model. This model may be based on popular deep learning architectures such as BERT or LSTM, which have been trained on large amounts of text data to accurately identify and evaluate the emotional tendency of words in the text. In practice, each word extracted from the text is fed into the emotion analysis model, which assigns each word an emotion score, typically within a range, such as-1 to 1, representing the emotion tendencies from negative to positive.

These emotion scores are then used to adjust the weight of each word in the keyword extraction. Specifically, words that are strongly related to positive emotions may be weighted more heavily, while words that are related to negative emotions may be weighted less heavily. This emotion-based weight adjustment helps the model more accurately identify words that are critical to understanding the user's attitudes and preferences for clothing.

Finally, the emotion weighted text features processed by the method provide input for subsequent feature selection and optimization mechanisms. The emotion analysis weighted features not only contain key information of the original text, but also incorporate emotion dimensions, so that the model is more accurate and sensitive when extracting and analyzing keywords about fashion trends of clothing.

Keyword weight adjustment mechanism 204 adds an important dimension to the dynamic semantic analysis model by introducing a deep learning emotion analysis model. The mechanism not only improves the accuracy of the model when extracting the keywords, but also enhances the understanding capability of the model on the emotion tendencies in the text, so that the model can better capture and reflect the perception and trend of the user on the clothing.

In the dynamic semantic analysis model in this embodiment, the feature selection and optimization mechanism 205 plays a vital role. Its main responsibility is to further optimize emotion weighted text features to ensure that the model can more accurately extract and analyze keywords about fashion trends of the garment.

Once the keyword weight adjustment mechanism 204 has completed emotion analysis of each word in the dynamic feature representation and adjusted their weights, the emotion weighted text features are sent to feature selection and optimization mechanism 205 for processing. The core task of this mechanism is to identify the most contributing features to the model from a large number of features, while removing those unimportant or redundant information. This step is critical to improving the accuracy and efficiency of the model.

The implementation of feature selection and optimization mechanisms relies primarily on gradient-based feature selection methods. This method determines the importance of each feature by assessing its extent of influence on the model output. In practice, a back-propagation algorithm, for example, is typically used to calculate the gradient of each feature to the model predictions. These gradient values provide direct information about the importance of the feature: a higher gradient value means that the feature has a greater impact on the model output and is therefore more important.

Based on these gradient values, the feature selection and optimization mechanism may dynamically adjust the weights of the features and even decide whether to retain or reject certain features. This process is typically performed in multiple iterations, each of which adjusts the selection and weight of features based on new data and feedback. This not only optimizes the feature set, but also enables the model to adapt to new data and trends, thereby improving its performance over time.

Upon completion of feature optimization, the optimized text features are ready to be fed into the data fusion module 206. At this stage they will be combined with other types of data (e.g. clothing sales, price change data) for a more comprehensive and thorough analysis.

The feature selection and optimization mechanism 205 fine-tunes and optimizes feature sets using a gradient-based approach, making the dynamic semantic analysis model more accurate and efficient in extracting keywords on clothing popularity trends. The implementation of this process is critical to improving the overall performance and accuracy of the model.

In this embodiment, the data fusion module 206 plays a vital role in integrating various data obtained from different sources, providing a more comprehensive view of the final extraction of keywords on fashion trends of clothing. The following is a detailed implementation description of the data fusion module:

The main task of the data fusion module 206 is to integrate data from different sources, including optimized text features, clothing sales data, and price change data. This step is critical to providing a comprehensive trend analysis because it not only relies on information in the text data, but also incorporates actual market dynamics data.

First, the module receives text features optimized by feature selection and optimization mechanism 205. These features have been processed through emotion weighting and gradient-based feature selection methods, and thus they contain profound insights into the topics of clothing in text data. In addition, the module receives clothing sales volume data and price change data, which are directly from the market and reflect the purchasing behavior of consumers and the supply and demand of the market.

Next, the data fusion module integrates these different types of data together. This process involves converting non-text data (such as sales and price change data) into a format compatible with the text features. Typically, this involves normalization of the data, normalization, and conversion of the data into feature vectors that can be processed by a machine learning model using specific encoding or embedding techniques.

Once all the data is transformed and normalized, the data fusion module analyzes the fused data using advanced analysis techniques, such as ensemble learning methods (e.g., random forests or gradient lifts). Through this analysis, the module can identify potential associations and patterns between different data sources and extract the most valuable information, which is critical to understanding and predicting fashion trends of clothing.

Finally, the result of the comprehensive analysis is an exhaustive list of keywords relating to fashion trends of clothing. These keywords are not only based on semantic analysis in text data, but also incorporate actual market data, thereby providing data-driven, comprehensive insight into fashion trends for apparel brands and retailers.

In summary, the data fusion module 206 is an integral part of the dynamic semantic analysis model. By fusing various types of data together and applying advanced analysis technology, the method provides a comprehensive and multi-dimensional view for capturing and understanding fashion trends of clothing.

The reference implementation code of the dynamic semantic analysis model is as follows

/>

In the above code, the implementation of each module is conceptual. For example, the preprocessing module uses TfidfVectorizer for simple text cleansing and word embedding; the local feature extraction module and the sequence feature extraction module respectively use simple CNN and LSTM network structures; the keyword weight adjustment mechanism, the feature selection and optimization mechanism, and the data fusion module all comprise basic example logic.

In practical applications, it is necessary to customize these modules according to data characteristics and traffic requirements, and to select appropriate neural network architecture and parameters.

For training a dynamic semantic analysis model, the following steps need to be followed:

1. Preparing training data:

a large amount of text data about the clothing is collected, such as user ratings, social media posts, etc.

These text data are preprocessed, including cleaning, normalization, and word embedding.

2. Defining a model architecture:

A Convolutional Neural Network (CNN) module is created to extract local features in the text data.

A Recurrent Neural Network (RNN) module is created to process the sequence data, capturing long-term dependencies and context information.

A pre-trained emotion analysis model (e.g., BERT or LSTM based model) is integrated for keyword weight adjustment.

The design feature selection and optimization mechanism may include a gradient-lifting decision tree or other advanced machine learning algorithm.

And a data fusion module is constructed for integrating text characteristics and other related data, such as sales and price changes.

3. Writing a loss function and selecting an optimizer:

A loss function, such as cross entropy loss, is defined for evaluating the performance of the model.

An optimizer, such as Adam or SGD, is selected for the training process of the model.

4. Training a model:

the model is trained using the prepared training data.

In the training process, local features are extracted through the CNN module, and sequence features are extracted through the RNN module.

Keyword weights are adjusted using emotion analysis models.

Feature selection and optimization mechanisms are applied to optimize the input features of the model.

Different data sources are integrated in the data fusion module.

Loss and accuracy in the training process are monitored, and model parameters or architecture are adjusted as required.

5. Model evaluation and tuning:

the performance of the model is evaluated using the validation dataset.

And adjusting a model framework or training parameters according to the evaluation result so as to improve the accuracy and generalization capability of the model.

6. Model test:

The model is run on the test dataset to ensure its accuracy and stability.

And outputting an analysis model to ensure that the keywords accurately reflect the fashion trend of the clothing.

Note that model training is an iterative process that may require multiple adjustments and optimizations to achieve optimal performance.

In summary, step S102 converts a large amount of unstructured text data into valuable insight regarding fashion trends of clothing by utilizing advanced natural language processing techniques. This step is not only critical to understanding the current market dynamics, but also provides critical support for subsequent predictive analysis and personalized recommendations.

Step S103: and analyzing the keywords related to the fashion trends of the clothing to identify the rising or falling trend of the fashion elements, and predicting the fashion trends of the clothing in the future according to the identified rising or falling trend of the fashion elements.

In step S103, the present embodiment adopts a detailed analysis method to process the keywords extracted in step S102 regarding the fashion trends of the clothing. This step is the core of this embodiment, which not only involves understanding of current market trends, but also concerns with predicting future trends, which is critical to achieving accurate garment recommendations.

First, this step focuses on deep analysis of the extracted keywords. Such analysis aims at identifying the trend of variation of popular elements implied in these keywords, such as whether the popularity of a certain style, color or material is increasing or decreasing. To achieve this, a range of data analysis techniques may be employed, including but not limited to time series analysis, trend line analysis, and pattern recognition algorithms. By these methods, the system is able to identify significant trends and patterns from historical and current data.

Next, the present embodiment further uses these analysis results to predict future fashion trends of clothing. Such predictions are based on understanding past and present data trends, combined with statistical models and predictive algorithms, such as linear regression, machine learning models, and the like. The ability to predict future trends is critical to maintaining the relevance and look-ahead of clothing recommendations. For example, by identifying that a clothing element is becoming a popular trend, the system may recommend related products in advance before the element becomes the main stream.

Furthermore, the analysis and prediction in step S103 is not a one-time activity, but a continuous process. The system repeats this step periodically to ensure that understanding of the popularity trends remains up to date. This is particularly important for the fashion industry to accommodate rapid changes, as the fashion trends can change significantly in a short period of time.

The analyzing the keywords about the fashion trend of the clothing to identify the rising or falling trend of the fashion element, and predicting the fashion trend of the clothing in the future according to the identified rising or falling trend of the fashion element, including:

In this embodiment, the calculating the trend score of the keyword based on the occurrence frequency and the growth rate of the keyword about the fashion trend of the clothing includes:

Wherein F _k is the frequency of occurrence of the keyword k, F _k divides the total number of words by calculating the number of occurrences of the keyword within a specified time window; the number of occurrences of the keyword k within a specified time window refers to the total number of occurrences of the keyword k within a selected time range (such as one week, one month, or one quarter). For example, if the term "environmental materials" is being studied and is mentioned 100 times in all relevant data sources within a month, then this value is 100.

Total word number: this generally refers to the total number of words in all text data within the same time window. This total word number includes all words, not just the keyword k, for example, if 50,000 words are analyzed in total in one month of data, then this number is 50,000.

SS _k is the emotion score of keyword k, and SS _k is calculated as the number of positive comments minus the number of negative comments divided by the total number of comments referring to the keyword;

Alpha, beta and gamma are weight parameters which can be directly set according to expert knowledge or obtained according to experimental data; GR _k is the growth rate of the keyword k, and GR _k is calculated using the following formula 2:

For each keyword extracted in step S102, the frequency of occurrence thereof within a given time window and the growth rate compared with the previous same time window are calculated. The frequency represents the popularity of the keyword, and the rate of increase reflects the rate of change of its popularity trend.

And classifying the keywords by using a clustering algorithm (such as K-means or hierarchical clustering). This step aims at classifying keywords with similar characteristics into the same category, forming keyword groups representing different flow trends.

The K-means clustering algorithm is described below as an example. K-means is a widely used clustering method that groups data points by assigning the data points to nearest cluster centers to form clusters. The following is a brief description of K-means clustering based on keyword trend scores:

Step 1, data preparation:

1. Collecting data: first, keyword data about garments is collected and preprocessed.

2. Calculating a trend score: trend scores are calculated for each keyword, possibly based on factors such as frequency of occurrence, growth rate, emotion analysis, etc.

3. Constructing a feature vector: the trend score for each keyword is converted into a feature vector. In the simplest case, this vector may contain only the trend score itself, but may also contain other relevant information, such as historical changes in the score.

Step 2, executing K-means clustering:

1. Selecting the number of clusters K: the number of clusters to be formed is determined. This may require estimating the optimal number of clusters based on traffic demand or by methods such as elbow rules.

2. Initializing a clustering center: k data points were randomly selected as initial cluster centers.

3. Assigning data points to nearest cluster centers: for each keyword, it is assigned to the nearest cluster center according to its feature vector. This is typically calculated based on euclidean distance.

4. Updating a clustering center: the center point of each cluster is recalculated as the average of all the feature vectors of the data points within the cluster.

5. Repeating the iteration: the steps of distributing and updating are repeated until the cluster center is not changed any more or the preset iteration times are reached.

And 3, analyzing and applying clustering results:

1. And (5) analyzing and clustering: each cluster represents a set of keywords with similar trend scores, possibly indicating a particular clothing trend.

2. And (5) extracting insight: based on the clustering results, it can be identified which types of clothing elements or styles are currently popular, which are becoming popular, or which are losing popularity.

In each cluster category, the keyword with the highest trend score is selected as a representative of that category. These representative keywords are considered to most accurately reflect the popularity trends of the category to which they belong.

For each representative keyword, a time series model (e.g., ARIMA, seasonal ARIMA, or long and short term memory network LSTM) is built. These models use historical trend scores of keywords to predict future trends. And predicting trend scores of representative keywords in a certain future time range by using the established time sequence model, so as to predict the future epidemic trend.

The following description will take the ARIMA model as an example:

Step one: data preparation

Trend scoring data is collected over a period of time for each representative keyword. These data may be recorded weekly or monthly.

Step two: establishing ARIMA model

A time series analysis was performed on the historical trend scores for each representative keyword using ARIMA model.

The ARIMA model requires the determination of three parameters: autoregressive term (AR), differential times (I), and moving average term (MA). These parameters can be determined by observing the autocorrelation and partial autocorrelation maps of the data.

Step three: predicting future trends

Using the ARIMA model, trend scores for each representative keyword can be predicted for the next months or even quarters. These predictions will reflect the likely direction of future fashion trends.

The predicted trends of all representative keywords are combined to form a prediction of the overall fashion trend of the garment. This may be achieved by weighted averaging or other statistical methods to ensure that the effects of the different trend categories are properly reflected.

A more specific example is provided below, assuming a hypothetical scenario in which keyword clustering and single trend prediction have been performed according to previous steps. Now, it is necessary to integrate this information to predict overall fashion trends.

The data analysis and prediction process is assumed to produce three main trend categories, each with a representative keyword and its predicted trend score:

1. Trend category a- "sustainable fashion":

-representative keywords: environment-friendly material "

Trend score prediction for 6 months into the future: 5.2,5.4,5.6,5.8,6.0,6.2

2. Trend category B- "antique style":

-representative keywords: ancient design "

Trend score prediction for 6 months into the future: 4.5,4.7,4.8,5.0,5.1,5.3

3. Trend category C- "scientific apparel":

-representative keywords: intelligent fabric "

Trend score prediction for 6 months into the future: 3.8,4.0,4.2,4.4,4.6,4.8

Comprehensive prediction of epidemic trend:

Step one: calculating an average trend score for each category:

For each trend category, an average trend score was calculated for the next 6 months. For example, for the "sustainable fashion" category, the average score would be (5.2+5.4+5.6+5.8+6.0+6.2)/6=5.7.

Step two: determining weights:

Assuming that based on market research and user research data, the following weights were determined: sustainable fashion (40%), antique style (35%), scientific apparel (25%). This reflects the market impact and user attention of each trend category.

Step three: calculating a weighted average trend score:

A weighted average score for the overall trend is calculated using the determined weights. For example, the weighted average score would be 5.7 x 0.40+4.9 x 0.35+4.3 x 0.25 (assuming 4.9 and 4.3 are the average scores of the other two categories, respectively).

The resulting weighted average score will represent the popularity trend for the entire apparel industry for the next 6 months. This scoring may help garment designers, retailers, and market analysts to understand which elements may be the focus of attention for the consumer in the future.

This score is calculated based on the various trend categories and their market impact or user attention, so it represents the overall direction of the overall market trend. To derive a specific hole and application from this score, the following considerations need to be combined:

(1) Interpreting representative keywords for each trend category:

The weighted average score is calculated based on the predictive scores of the representative keywords for each trend category. Thus, understanding representative keywords for each category (e.g., "environmental materials," "antique designs," "smart fabrics") will help us know which specific popular elements in the market are gaining attention.

(2) Weights for trend categories are considered:

the weight of each trend category when calculating the weighted average score reflects its market impact or user attention. For example, if "sustainable fashion" gets a higher weight, this means that the market is rising for environmentally friendly products.

(3) Comprehensive analysis and market policy establishment:

By combining scoring analysis with market research, consumer behavior analysis, and fashion trend prediction, clothing designers, retailers, and market analysts can identify those elements or styles that may be future hotspots and adjust product design, inventory management, and marketing strategies accordingly.

(4) Dynamic monitoring and adjustment:

As market and consumer preferences change, the importance of the trend may change. Periodically repeating trend analysis and scoring calculations can help related personnel maintain sensitivity to market dynamics and adjust strategies in time to accommodate new trends.

It can be seen that the weighted average score provides a quantified, data-based perspective to understand the overall direction of market trending. By deep analysis of this score and the underlying trend categories, industry professionals can gain valuable insight into future popularity trends.

By this means, not only is the development of each individual trend category predicted, but the overall popularity direction of the whole apparel industry can be understood. Such comprehensive analysis is critical to developing effective marketing strategies and meeting future consumer needs.

The following is another analysis method of fashion trends of clothing, which can be used as a supplement to this embodiment.

In the apparel industry, popular trend analysis is a critical task that involves comprehensive statistics, analysis and summarization of current and future popular apparel styles. This includes careful examination of various aspects of the garment, such as the contours, parts, fabric, colors, patterns, and processes. The following is a technical solution on how to perform a popular trend analysis.

First, the profile is analyzed. The profile is the overall contour and shape of the garment, which is an important factor in determining the style of the garment. By analyzing the wear of different fashion shows, fashion magazines, fashion blouse and fashion collars, the currently popular profile can be identified. Such as whether a loose shirt, tight skirt, or a fluffy skirt is popular, etc.

Second, component analysis is also critical. The parts comprise various components of the garment, such as necklines, sleeves, buttons, etc. By looking at the designs that are popular in the market, it is possible to see which types of collar or sleeve designs are currently popular.

Fabric analysis involves examining the trend of use of different materials. This may include analyzing which fabrics are popular in the season, such as light and thin breathable cotton, gorgeous silk, or comfortable knitted fabrics, etc. In addition, there is also a concern about the trend of sustainable materials, which is particularly important in modern fashion.

Color and pattern analysis involves examining the colors and patterns currently in popular use. It is possible to determine which colors or patterns are likely to be the next popular trend by analyzing the colors and patterns of fashion shows, popular color predictions, and popular products on the market.

Finally, the technology and the technique adopted for making the clothing are examined by technological analysis. This includes understanding of different sewing techniques, decorative details or unique manufacturing processes, and how they affect garment designs and fashion trends.

To perform these analyses, a variety of methods may be employed. First, data from a variety of channels such as fashion shows, fashion weeks, social media, fashion magazines, etc., are collected and consolidated, which are the primary sources of understanding the latest fashion trends. Data analysis tools, such as data mining and image recognition software, are then used to automatically identify and classify the different design elements. These tools can help quickly analyze large amounts of picture and text data to identify popular elements such as profiles, colors, patterns, etc.

In addition, communication with professionals in the fashion industry is also an important approach, and their insight and predictions can provide valuable information for trend analysis. At the same time, utilizing consumer research and market research data can also provide important insights about consumer preferences and behavior.

In summary, popular trend analysis is a complex but orderly process involving the investigation of apparel designs from multiple angles and levels. By adopting a data-driven method and combining industry expert knowledge, the future popularity trend can be effectively predicted and analyzed, and guidance is provided for clothing design and market strategies.

In summary, step S103 is a crucial step in the embodiment, which involves not only complex analysis of large amounts of data, but also requires a high degree of flexibility and adaptability of the system for accurate prediction and timely updating of trends. Through the step, the fashion trend of the clothing can be captured and predicted, and a solid foundation is laid for subsequent personalized recommendation.

Step S104: based on the historical purchase data and browsing behaviors of the user and the predicted fashion trends of the clothing, a personalized clothing recommendation list is generated.

In step S104, a personalized clothing recommendation list is generated using the analysis results of the previous steps. The implementation of this step is critical because it directly affects the personalization and accuracy of the end user experience.

First, this step involves in-depth analysis of the user's historical purchase data and browsing behavior. This includes analyzing information of the user's past shopping preferences, types of goods frequently viewed, purchase cycles, etc. These data provide valuable insight into the personal preferences and purchasing habits of each user. This understanding is the cornerstone of the personalized recommendation system, as it allows the system to customize the more relevant and attractive clothing options for each user.

Subsequently, the system further refines the recommendation list in combination with the fashion trends of the clothing predicted in step S103. This means that the recommendation is not only based on the past behavior of the user, but also incorporates the latest market trend. For example, if a user prefers a recreational style and a certain color of recreational styles currently on the market is very popular, the system may prefer to have the user recommended clothing of that color. This approach of combining historical data with current trends allows recommendations to be both personalized and fashionable.

In addition, step S104 also considers technical details of implementing personalized recommendations. This typically involves complex algorithms and machine learning models, such as collaborative filtering, content recommendation algorithms, and the like. These models are able to learn and identify patterns from a large amount of user data, thereby generating the most appropriate clothing recommendations for each user.

To ensure continuous optimization and updating of the recommendation list, this step may also include a dynamic feedback mechanism. The recommendation algorithm is adjusted and optimized based on the actual user response to the recommended clothing, such as click through rate, purchase rate, and feedback. This means that over time the recommendation will become more accurate and personalized.

Finally, step S104 is a key step in the embodiment to achieve end user value. Through the step, clothing choices which reflect personal tastes and follow popular trends can be provided for each user, so that user satisfaction and shopping experience are greatly improved.

In this embodiment, the generating a personalized clothing recommendation list based on the historical purchase data and browsing behavior of the user and the predicted clothing popularity trend includes:

First, historical purchase data and browsing behavior of the user need to be collected. The historical purchase data primarily includes the type of clothing and brands that the user purchased in the past, which may be obtained by analyzing the user's purchase history on the e-commerce platform. Such data provides direct evidence about user preferences, such as that a user may prefer to purchase casual style clothing or a particular brand of product. At the same time, data of browsing behavior is also collected, including the history of pages browsed by the user on the e-commerce platform. Such data may show which products are of interest to the user, which products they spend more time viewing, and may even capture their interest in not yet purchased goods.

Next, based on the user's historical purchase data and browsing behavior, the method involves identifying the user's preferences for particular styles, and categories of clothing. This step may involve using data analysis and machine learning techniques, such as pattern recognition and user behavior analysis, to parse and understand the user data. For example, if a user often purchases a garment of a certain color or style, the system may infer that the user may have particular preferences for that color or style.

Finally, the user's preferences for specific styles, styles and categories of clothing, as derived by the analysis, are combined with current and predicted fashion trends of clothing. This means that not only personal preferences of the user but also popular trends in the market are taken into account in order to generate a list of clothing recommendations that meets personal tastes and does not deviate from fashion trends. This fusion process may involve complex data fusion and analysis, ensuring that the recommendation results are both personalized and fashionable.

Through the steps, the method can provide the personalized clothing recommendation list which considers personal preferences of users and reflects popular trends for the users. Such recommendations not only help to improve user satisfaction and shopping experience, but may also help the e-commerce platform to improve sales efficiency and customer loyalty.

In this embodiment, the collecting historical purchase data and browsing behavior of the user includes:

First, the login information and browser cookie data of the user on the e-commerce platform are utilized. When a user logs onto an e-commerce platform, the system may identify the user's identity through the login information. Meanwhile, the system can track the browsing behavior of the user through the cookie data of the browser. Such tracking may include the pages of the merchandise that the user browses, the time spent on a particular page of merchandise, the user's search history, and their clicking actions on the merchandise. In addition, if the user makes purchases on the platform, data for those purchases may also be collected. This includes information on the type of goods purchased by the user, the time of purchase, the frequency of purchase, etc.

All of this collected data is then stored in the user's profile. This includes not only the browsing behavior data of the user, but also their purchase history. Storing these data is critical for subsequent data analysis and personalized recommendations. For example, by analyzing the user's purchase history, the user's preferences for certain brands or styles of apparel may be found; by analyzing the browsing behavior of the user, it is possible to find out which products the user shows a higher interest in.

This data collection and storage process provides the basis for the next steps. Based on these data, various data analysis techniques and machine learning algorithms may be used to analyze the user's purchase and browsing patterns to generate a personalized clothing recommendation list. The recommendation list not only considers the historical behaviors of the user, but also can combine the current fashion trend of clothing, thereby providing the user with clothing choices which are in line with personal tastes and fashion.

In this embodiment, the identifying the user's preference for a specific clothing style, style and category according to the user's historical purchase data and browsing behavior includes:

First, a browsing behavior of a user on an e-commerce platform is analyzed. In particular, the system may analyze the user's dwell time on different merchandise pages, the user's click frequency on a particular merchandise, and the user's search history. For example, if a user has a long dwell time on a certain type of clothing page or frequently clicks to view details of such clothing, this may indicate that the user is interested in this type of clothing. Likewise, the user's search history may also reveal their preferences for a particular style, color, or brand. Together, these data points help the system identify the intensity of interest of the user in different products.

Next, a deep analysis is performed on the user's purchase history. This step aims to determine the brand, style and price preferences of users by evaluating their past purchasing behavior. For example, by analyzing the brand and style of apparel that the user purchased in the past, the system can identify what types of products the user tends to purchase. In addition, the frequency of purchases by the user and the purchasing behavior within a particular price range also provide important clues to determining the user's consumption habits and price sensitivity.

By combining these two analysis: i.e., browsing behavior and purchase history of the user, the method can comprehensively evaluate clothing preferences of the user. This comprehensive analysis helps the system more accurately understand the personal tastes of each user, thereby generating a more personalized list of recommendations. These recommendations are not only based on the actual purchasing behavior of the user, but also take into account the interests and preferences they exhibit during browsing.

To implement this process, the technician needs to employ data analysis and machine learning techniques. Data analysis techniques may be used to parse the user's browsing and purchasing data, while machine learning algorithms may help identify and predict the user's preference patterns.

In general, by the method, more accurate and personalized clothing recommendation can be provided for the user, so that user satisfaction and shopping experience are improved.

In this embodiment, the step of combining the identified preference for the specific clothing style, style and category with the predicted fashion trend of clothing to generate the clothing recommendation list for the user includes:

This process involves first identifying the user's personal preferences, which includes analyzing the user's historical purchase data and browsing behavior to determine their preferences for particular styles, styles and categories. These preferences may relate to a particular brand that the user is inclined to purchase, color preferences, price ranges, or their continued purchasing behavior for a certain style.

Next, the method needs to consider the current fashion trends of clothing. This is typically accomplished by analyzing market data and fashion trend predictions.

Once both personal preferences and market trends are determined, the next step is to generate a personalized recommendation list using collaborative filtered recommendation algorithms. Collaborative filtering is a commonly used recommender algorithm that predicts new items that a user may be interested in based on the user's past behavior and other similar user behaviors. In this case, the algorithm will take into account the user's personal preferences and overall market trends and then recommend those garments that meet the user's preferences and fashion.

By combining the personal preference and market trend of the user and applying advanced recommendation system technology, the method can provide the user with clothing choices which are in line with personal tastes and keep up with fashion trends, thereby improving user satisfaction and shopping experience.

A second embodiment of the present application provides an electronic apparatus including:

A processor;

and a memory for storing a program which, when read and executed by the processor, performs the big data based clothing popularity trend capturing and recommending method provided in the first embodiment of the present application.

A third embodiment of the present application provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the big data based fashion trend capturing and recommending method provided in the first embodiment of the present application.

While the application has been described in terms of preferred embodiments, it is not intended to be limiting, but rather, it will be apparent to those skilled in the art that various changes and modifications can be made herein without departing from the spirit and scope of the application as defined by the appended claims.

Claims

1. A method for capturing and recommending fashion trends of clothing based on big data, comprising:

A personalized clothing recommendation list is generated based on historical purchase data and browsing behavior of the user and predicted future clothing popularity trends.

2. The big data based clothing popularity trend capturing and recommending method of claim 1, wherein the analyzing the clothing-related text data using natural language processing technology to extract keywords on clothing popularity trends comprises:

Inputting the clothing related text data into a trained dynamic semantic analysis model to obtain keywords of clothing popular trends, wherein the dynamic semantic analysis model comprises a preprocessing module, a local feature extraction module, a sequence feature extraction module, a keyword weight adjustment mechanism, a feature selection and optimization mechanism and a data fusion module;

3. The method for capturing and recommending fashion trends of clothing based on big data according to claim 1, wherein the analyzing the keywords on fashion trends of clothing to identify rising or falling trends of fashion elements and predicting future fashion trends of clothing according to the identified rising or falling trends of fashion elements comprises:

4. The big data based clothing popularity trend capturing and recommending method of claim 3, wherein calculating a trend score for keywords based on the frequency of occurrence and the rate of increase of keywords with respect to clothing popularity trends, comprising:

5. The big data based clothing popularity trend capturing and recommending method of claim 1, wherein the generating a personalized clothing recommendation list based on historical purchase data and browsing behavior of the user and predicted future clothing popularity trends, comprises:

6. The big data based clothing popularity trend capturing and recommending method of claim 5, wherein the collecting historical purchase data and browsing behavior of the user comprises:

7. The big data based clothing popularity trend capturing and recommending method of claim 5, wherein identifying user preferences for specific clothing styles, styles and categories based on the user's historical purchase data and browsing behavior comprises:

8. The big data based clothing popularity trend capturing and recommending method of claim 5, wherein combining the identified preferences for specific clothing styles, and categories with the predicted clothing popularity trends generates a list of clothing recommendations for the user, comprising:

9. A feature generation apparatus of network traffic data, comprising:

A processor;

A memory for storing a program which, when read by the processor, performs the big data based clothing popularity trend capturing and recommending method provided in any one of claims 1-8.

10. A computer readable storage medium, having stored thereon a computer program which, when executed by a processor, performs the big data based clothing popularity trend capturing and recommending method provided in any of claims 1-8.