Fashion Conversation Data on Instagram

The fashion industry is establishing its presence on a number of visual-centric social media like Instagram. This creates an interesting clash as fashion brands that have traditionally practiced highly creative and editorialized image marketing now have to engage with people on the platform that epitomizes impromptu, realtime conversation. What kinds of fashion images do brands and individuals share and what are the types of visual features that attract likes and comments? In this research, we take both quantitative and qualitative approaches to answer these questions. We analyze visual features of fashion posts first via manual tagging and then via training on convolutional neural networks. The classified images were examined across four types of fashion brands: mega couture, small couture, designers, and high street. We find that while product-only images make up the majority of fashion conversation in terms of volume, body snaps and face images that portray fashion items more naturally tend to receive a larger number of likes and comments by the audience. Our findings bring insights into building an automated tool for classifying or generating influential fashion information. We make our novel dataset of 24,752 labeled images on fashion conversations, containing visual and textual cues, available for the research community. Social media has become an important platform for the fashion industry for testing new marketing strategies and monitoring trends (Kim and Ko 2012). Already thousands of luxury and high street fashion brands around the world are present online and communicate with their followers and potential customers (Hu et al. 2014). While fashion brands have unilaterally set their polished brand images through traditional media such as television channels and magazines, two unique properties of social media serve as a very powerful tool for promoting and sharing fashion information to both industry and people. Firstly, the interactive nature of social media allows anyone to generate content and participate in establishing brand images. Not only large fashion houses launch advertising campaigns and share their latest runway looks through social media platforms, individuals and local stores also contribute Copyright c © 2017, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. to the online fashion conversation by sharing purchase experiences or new trends. Individuals discuss and rate fashion products openly and favorable reviews spread via word-ofmouth. In contrast, dissatisfied customers leave harsh criticism and complaints on fashion products online. Secondly, many social media platforms are visual-centric and heavily utilize images and videos. Visuals, a powerful tool in advertising and communication (Messaris 1996; Joo et al. 2014), are critical in fashion marketing because appearance is the key information of any fashion look. In addition, compared to traditional platforms, social media offer much higher bandwidth in that brands can now deliver information about a single fashion product through hundreds of varying images. Among various platforms, Instagram has reshaped the fashion industry landscape. Plenty of fashion brands are hosting new marketing campaigns based on it’s hashtag functions and consumer-generated contents. To meet the spontaneous nature of the platform, some brands also adopt non-traditional photographic styles such as “behind the scenes” that are secondary, less-editorialized images to create unique brand stories. However, little is known about the fashion conversation itself. Lack of any labeled data describing fashion style is a barrier to investigating such trends. Building a comprehensive fashion dataset will enable research on customized recommendations that associate personal tastes with fashion picks as well as identify emerging trends from different parts of the world. Such research will enable fashion industry to better understand how products and brands are perceived by people. As a result building fashion datasets helps create new products in a sustainable way. According to the World Economic Forum, fashion is the second largest polluter of environment after oil and creating innovative ways to reduce fashion waste is a critical challenge. We envision to take on the challenge of analyzing how fashion tastes are shared and disseminated on social media. As a first step, we built a sizable yet detailed labeled dataset describing conversations on notable fashion brands on Instagram. By employing deep learning techniques, we identified meaningful topics in the context of fashion and automatically labeled the gathered images. This paper presents the steps involved in the labeling task and shares the data for further discussion. Our contributions are as follows: ar X iv :1 70 4. 04 13 7v 1 [ st at .M L ] 1 3 A pr 2 01 7 1. We release a novel dataset describing 24,752 fashion images of 48 brands on Instagram with meaningful visual tags. We identified five major visual categories of fashion images via comprehensive content labeling, which are selfie, body snap, marketing shot, product-only, and nonfashion. We also trained a convolutional neural network (CNN) to classify the major visual categories and other important visual features from images including ‘face’ or ‘brand logo.’ 2. Our analysis on visual content of fashion images and audience engagement reveals an interesting discrepancy between post volume and reactions; while product-only images are the most common in terms of volume, body snaps and photos containing faces that reveal fashion items more naturally receive disproportionately large number of likes and comments from the audience (e.g., 31% of the fashion posts receiving 53% of total likes). 3. We identify challenges in fashion image classification; we encounter a number of non-canonical images that do not fall into the five major categories such as advertisements, clickbaits, zoom-in shots of textile and products, and multi-functional images. These images nonetheless explicitly contained fashion hashtags. 4. Regression and ANOVA tests indicate what kinds of image features and emotions draw more attention from audience. We find that body snap and face features are a better communicator than product-only or logo features and that certain facial expressions like happiness and neutral emotion show a significant relationship with the likes count. Our findings bring theoretical and practical implications for studying fashion conversations on social media. The tagged images can be used for defining what images are considered in the domain of fashion in user generated content. For example, the practice of sharing images containing faces with fashion hashtags needs to be understood better. Some of these images dedicated more visual space to the face itself than the associated fashion items (i.e., selfie shots where faces take up more than half of photo length). We also observed a number of image spams that exploit irrelevant fashion tags, and hence our data can be potentially used for identifying fashion clickbaits. Finally our data indicate that product-only images dominate the conversation of most fashion brands, yet images of this type are less effective in gaining likes and comments. Such information will be useful for fashion brands and marketers to effectively promote their products and communicate their brand images to the customers.