ArticlePublisher preview available

Multi-view Latent Learning Applied to Fashion Industry

Authors:
To read the full-text of this research, you can request a copy directly from the authors.

Abstract and Figures

Demand forecasting is one of the main challenges for retailers and wholesalers in any industry. Proper demand forecasting gives business valuable information about potential profits and helps managers in taking targeted decisions on business growth strategies. Nowadays almost all organizations use different data sources or databases for nearly every aspect of their operations so that the knowledge on products on sale belongs to several independent views. The methodology described in this paper addresses the issue of product demand forecasting in fashion industry exploiting a multi-view learning approach. In particular, we show how the integration and connection among multiple views improves results accuracy. In real-life applications not all the views are usually available before a product is put on the market but the utility of a proper demand forecasting increases if the prediction is available before the product launch. We show that missing views can be reconstructed by means of common latent factors; in particular, this paper presents a learning procedure that describes the connection between different views. This connection allows data integration from multiple sources and can be extended to the special case of partial data representation. The nearest neighbors in the latent space play a special role for this process and for a general improvement of the forecast quality. We experimented the proposed methodology on real fashion retail sales showing that multi-view latent learning provides a system that is able to reconstruct satisfactorily non yet available views and can be used to predict the volumes of sales well before the goods are put on the market.
This content is subject to copyright. Terms and conditions apply.
https://doi.org/10.1007/s10796-020-10005-8
Multi-view Latent Learning Applied to Fashion Industry
Giovanni Battista Gardino1·Rosa Meo2·Giuseppe Craparotta3
©Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract
Demand forecasting is one of the main challenges for retailers and wholesalers in any industry. Proper demand forecasting
gives business valuable information about potential profits and helps managers in taking targeted decisions on business
growth strategies. Nowadays almost all organizations use different data sources or databases for nearly every aspect of their
operations so that the knowledge on products on sale belongs to several independent views. The methodology described in
this paper addresses the issue of product demand forecasting in fashion industry exploiting a multi-view learning approach.
In particular, we show how the integration and connection among multiple views improves results accuracy. In real-life
applications not all the views are usually available before a product is put on the market but the utility of a proper
demand forecasting increases if the prediction is available before the product launch. We show that missing views can be
reconstructed by means of common latent factors; in particular, this paper presents a learning procedure that describes the
connection between different views. This connection allows data integration from multiple sources and can be extended to
the special case of partial data representation. The nearest neighbors in the latent space play a special role for this process
and for a general improvement of the forecast quality. We experimented the proposed methodology on real fashion retail
sales showing that multi-view latent learning provides a system that is able to reconstruct satisfactorily non yet available
views and can be used to predict the volumes of sales well before the goods are put on the market.
Keywords Multi-view learning ·Latent spaces ·Fashion forecast
1 Introduction
Demand forecasting is one of the main challenges for
retailers and wholesalers in any industry. Proper demand
forecasting gives businesses valuable information about
potential profits in their current market so that managers
can take targeted decisions on pricing and business growth
strategies. If demand is underestimated, future sales can be
lost; on the other side, if suppliers are left with a surplus,
heavy markdowns strategies could be forced, implying
possible losses and cash flow issues.
In fashion industry, demand forecasting is more than
that: seasonal trends, lack of data and general uncertainty
make the prediction often difficult and unstable. A good
Rosa Meo
meo@di.unito.it
1GDP Analytics, Torino, Italy
2University of Torino, Torino, Italy
3EVO Pricing, Torino, Italy
fashion forecaster must take into account lots of variables,
such as political and economical system, demographics
of certain areas, consumer expectation, market trends,
internal business strategies and many others. In this context,
“forecasting” has two main goals: projecting past trends into
the future and anticipating forthcoming developments by
looking for signs of change.
The difficulty and uncertainty of the problem has caused
many debates over the last decades. On one hand people are
skeptical and believe that trends forecast may be regarded as
accurate because of its self-fulfilling process (see Adegeest
2016); on the other hand fashion analysts ensure that big
data and machine learning algorithms are becoming every
day more relevant (see Erhard and Bug 2016).
Recent researches by Na et al. (2013), Thomassey
(2010), and Thomassey (2014) show how advanced
analytical techniques have become an important part in
one of the most intuition-based and unpredictable indstry:
although, the use of big data cannot ever completely
redefine fashion industry as it is more an art than a
science, it definitely is revolutionizing the way industrialists
and brands produce apparels and accessories. Most of
Published online: 4 April 2020
Information Systems Frontiers (2021) 23:53–69
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
... efficient data models, data processing pipelines and architectures to integrate standard and big data sources (Jovanovic et al. 2020) as well as to improve resource utilization and aggregate performance in shared environments (Michiardi et al. 2020); predictive analytics to forecast product demand in the fashion industry (Gardino et al. 2020) and techniques to deal with the lack of annotated data for sensor-based human activity recognition (Prabono et al. 2020); text data processing to assess the performance of text storage systems through a generic benchmark (Truicȃ et al. 2020) and innovative solutions to deal with specific use cases such as the legal domain (Bordino et al. 2020); novel approaches for mining social media to support intelligent transportation systems (Vallejos et al. 2020) and digging deep the IoT scenario (Ustek-Spilda et al. 2020); -solutions to deal with privacy issues in distance learning systems (Preuveneers et al. 2020). ...
... In this context, paper (Gardino et al. 2020) proposes a method for predicting product demand in the fashion industry. The proposed prediction method, called multi-VIew Bridge Estimation (VIBE), takes advantage of the existence of multiple views on items, i.e., sets of homogeneous features. ...
Chapter
Many problems can be correctly tackled only when the related information is available from various viewpoints. Thus, when a problem is to be solved by machine, the data from multiple views needs to be well presented to a machine learning algorithm that suitably processes and learns. For a machine learning algorithm to be able to process such data, finding complementary information between view pairs is important. In this paper, a similarity distance-based canonical correlation analysis (SDCCA) has been proposed to determine the complementary information by finding the similarity between view pairs for multiview data representation. The proposed approach uses the Bhattacharya similarity distance. It is evident from experimental results based on real-world datasets that the SDCCA approach performs better compared to existing canonical correlation analysis-based approaches. Thus is a more effective and promising approach for solving real-life problems where consideration of complementary information in multiple views is essential.Gupta, SurendraThakar, UrjitaTokekar, Sanjiv
Book
Full-text available
Like many others, fashion companies have to deal with a global and very competitive environment. Thus companies rely on accurate sales forecasts - as key success factor of an efficient supply chain management. However, forecasters have to take into account some specificities of the fashion industry. To respond to these constraints, a variety of different forecasting methods exists, including new, computer-based predictive analytics. After the evaluation of different methods, their application to the fashion industry is investigated through semi structured expert interviews. Despite several benefits predictive analytics is not yet frequently used in practice. This research does not only reflect an industry profile, but also gives important insights about the future potential and obstacles of predictive analytics.
Conference Paper
Full-text available
What is the future of fashion? Tackling this question from a data-driven vision perspective, we propose to forecast visual style trends before they occur. We introduce the first approach to predict the future popularity of styles discovered from fashion images in an unsupervised manner. Using these styles as a basis, we train a forecasting model to represent their trends over time. The resulting model can hypothesize new mixtures of styles that will become popular in the future, discover style dynamics (trendy vs. classic), and name the key visual attributes that will dominate tomorrow’s fashion. We demonstrate our idea applied to three datasets encapsulating 80,000 fashion products sold across six years on Amazon. Results indicate that fashion forecasting benefits greatly from visual analysis, much more than textual or meta-data cues surrounding products.
Article
Full-text available
The plenty information from multiple views data as well as the complementary information among different views are usually beneficial to various tasks, e.g., clustering, classification, de-noising. Multi-view subspace clustering is based on the fact that the multi-view data are generated from a latent subspace. To recover the underlying subspace structure, the success of the sparse and/or low-rank subspace clustering has been witnessed recently. Despite some state-of-the-art subspace clustering approaches can numerically handle multi-view data, by simultaneously exploring all possible pairwise correlation within views, the high order statistics is often disregarded which can only be captured by simultaneously utilizing all views. As a consequence, the clustering performance for multi-view data is compromised. To address this issue, in this paper, a novel multi-view clustering method is proposed by using \textit{t-product} in third-order tensor space. Based on the circular convolution operation, multi-view data can be effectively represented by a \textit{t-linear} combination with sparse and low-rank penalty using "self-expressiveness". Our extensive experimental results on facial, object, digits image and text data demonstrate that the proposed method outperforms the state-of-the-art methods in terms of many criteria.
Article
Full-text available
In this study, a clustering-based sales forecasting scheme based on support vector regression (SVR) is proposed. The proposed scheme first uses k-means algorithm to partition the whole training sales data into several disjoint clusters. Then, for each group, the SVR is applied to construct forecasting model. Finally, for a given testing data, three similarity measurements are used to find the cluster which the testing data belongs to and then employee the corresponding trained SVR model to generate prediction result. A real aggregate sales data of computer server is used as an illustrative example to evaluate the performance of the proposed model. Experimental results revealed that the proposed clustering-based sales forecasting scheme outperforms the single SVR without data clustering and hence is an effective alternative for computer server sales forecasting.
Article
Full-text available
Sales forecasting is crucial for many retail operations. It is especially critical for the fashion retailing service industry in which product demand is very volatile and product's life cycle is short. This paper conducts a comprehensive literature review and selects a set of papers in the literature on fashion retail sales forecasting. The advantages and the drawbacks of different kinds of analytical methods for fashion retail sales forecasting are examined. The evolution of the respective forecasting methods over the past 15 years is revealed. Issues related to real-world applications of the fashion retail sales forecasting models and important future research directions are discussed.
Article
Full-text available
We describe a completely automated large scale visual recommendation system for fashion. Our focus is to efficiently harness the availability of large quantities of online fashion images and their rich meta-data. Specifically, we propose four data driven models in the form of Complementary Nearest Neighbor Consensus, Gaussian Mixture Models, Texture Agnostic Retrieval and Markov Chain LDA for solving this problem. We analyze relative merits and pitfalls of these algorithms through extensive experimentation on a large-scale data set and baseline them against existing ideas from color science. We also illustrate key fashion insights learned through these experiments and show how they can be employed to design better recommendation systems. Finally, we also outline a large-scale annotated data set of fashion images (Fashion-136K) that can be exploited for future vision research.
Article
Multi-view learning is an emerging direction in machine learning which considers learning with multiple views to improve the generalization performance. Multi-view learning is also known as data fusion or data integration from multiple feature sets. Since the last survey of multi-view machine learning in early 2013, multi-view learning has made great progress and developments in recent years, and is facing new challenges. This overview first reviews theoretical underpinnings to understand the properties and behaviors of multi-view learning. Then multi-view learning methods are described in terms of three classes to offer a neat categorization and organization. For each category, representative algorithms and newly proposed algorithms are presented. The main feature of this survey is that we provide comprehensive introduction for the recent developments of multi-view learning methods on the basis of coherence with early methods. We also attempt to identify promising venues and point out some specific challenges which can hopefully promote further research in this rapidly developing field.
Chapter
The fashion industry is a very fascinating sector for the sales forecasting. Indeed, the long time-to-market which contrasts with the short life cycle of products, makes the forecasting process very challenging. A suitable forecasting system should also deal with the specificities of the demand: fashion trends, seasonality, influence of many exogenous factors, …. We propose here a review of the different constraints related to the sales forecasting in the fashion industry, the methodologies and techniques existing in the literature to cope with these constraints and finally, the new topics which could be explored in the field of the sales forecasting for fashion products.