Leveraging Multiple Relations for Fashion Trend Forecasting Based on Social Media

Abstract

Fashion trend forecasting is of great research significance in providing useful suggestions for both fashion companies and fashion lovers. Although various studies have been devoted to tackling this challenging task, they only studied limited fashion elements with highly seasonal or simple patterns, which could hardly reveal the real complex fashion trends. Moreover, the mainstream solutions for this task are still statistical-based and solely focus on time-series data modeling, which limit the forecast accuracy. Towards insightful fashion trend forecasting, previous work [1] proposed to analyze more fine-grained fashion elements which can informatively reveal fashion trends. Specifically, it focused on detailed fashion element trend forecasting for specific user groups based on social media data. In addition, it proposed a neural network-based method, namely KERN, to address the problem of fashion trend modeling and forecasting. In this work, to extend the previous work [1], we propose an improved model named Relation Enhanced Attention Recurrent (REAR) network. Compared to KERN, the REAR model leverages not only the relations among fashion elements, but also those among user groups, thus capturing more types of correlations among various fashion trends. To further improve the performance of long-range trend forecasting, the REAR method devises a sliding temporal attention mechanism, which is able to capture temporal patterns on future horizons better. Extensive experiments and more analysis have been conducted on the FIT [1] and GeoStyle [2] datasets to evaluate the performance of REAR. Experimental and analytical results demonstrate the effectiveness of the proposed REAR model in fashion trend forecasting, which also show the improvement of REAR compared to the KERN.

FashionReGen: LLM-Empowered Fashion Report Generation

“…Report the tagging results for each garment with a single line pattern, following the corresponding category. Do not output anything other than the category and tags for the mentioned 2 WGSN is professional fashion analysis company aspects. Here is an example for your tagging, <image: {Category: Top; Style: Layered, Modern, ... Silhouette: Relaxed, ...; ...}, {Category: Skirt; Style: Casual, Street, ...; ... }>.…”

“…The fashion industry is a vital component of global economy, characterized by its constant pursuit of novelty and changes. It is essential for fashion practitioners, enthusiasts and consumers to capture these shifts to get ahead of the right trends and make confident decisions [2][3][4]. To this end, specialized consulting and forecasting companies conduct seasonal or annual analysis, producing insightful fashion reports 1 .…”

A Review of Modern Fashion Recommender Systems

“…There is a large number of approved sizing systems around the globe for various clothes, such as dresses, tops, skirts, pants and brands. Moreover, there are different size systems such as numeric (38-39-40), standard (S,M,L), fractions (41 1/3, 42.5), convention sizes (36)(37)(38)(40)(41)(42), country conventions (EU, FR, IT, UK), where inconsistencies and different ways of converting a local size system to another (as brands do not always comply with the same conversion logic) make the task challenging. Subjectivity.…”

Leveraging Two Types of Global Graph for Sequential Fashion Recommendation

“…With the rapid development and tremendous success of GNNs in many application domains [2,15,25], recommendation approaches based on GNNs [5,33] have achieved state-of-the-art performance in various sub-tasks, such as implicit feedback-based general recommendation and session-based recommendation [10,30]. Graph is a natural data structure for most of data in recommender systems.…”

“…Implementation details: We perform grid search on the embedding size of users and items, which is within [5,10,20,50] . In terms of the number of propagation layers of the u-i graph and i-i graph, we perform grid search within [1,2,3] for each type of graph respectively. The learning rate during training is set to 0.01 and the batch size is set to 5000 for all methods.…”