The Rise of Fashion Informatics: A Case of Data-Mining-Based Social Network Analysis in Fashion

Abstract

With the advent of modern cognitive computing technologies, fashion informatics researchers contribute to the academic and professional discussion about how a large-scale data set is able to reshape the fashion industry. Data-mining-based social network analysis is a promising area of fashion informatics to investigate relations and information flow among fashion units. By adopting this pragmatic approach, we provide dynamic network visualizations of the case of Paris Fashion Week. Three time periods were researched to monitor the formulation and mobilization of social media users’ discussions of the event. Initial textual data on social media were crawled, converted, calculated, and visualized by Python and Gephi. The most influential nodes (hashtags) that function as junctions and the distinct hashtag communities were identified and represented visually as graphs. The relations between the contextual clusters and the role of junctions in linking these clusters were investigated and interpreted.

Keywords

network analysis data mining fashion informatics

Big data and advanced analytics are impacting the business dynamics of the fashion industry, yet current fashion researchers have not sufficiently taken advantage of large-scale data sets and data science (Lin, Zhou, & Xu, 2015). Traditionally, industry trends have followed a top-down process led simply by the inspiration and creativity of practitioners, which companies attempt to harness and then sell to the public. But the emotional aspects of creativity (practitioners) and preferences (consumers) are very hard to quantify and measure (Kindred & Steele, 2015). How to translate that creativity or intuition into a data-driven structure is a challenge for both fashion scholars and practitioners. With this in mind, a new interdisciplinary field is emerging called “fashion informatics,” which refers to the analysis of massive amounts of data by means of machine learning, social network analysis (SNA), and computer vision techniques targeted toward the fashion industry (Reiter, Zhao, Ciampaglia, & Ferrara, 2016). With the advent of modern cognitive computing technologies, fashion big data can be used in trends forecasting, influencer analysis, supply chain management, and personalized recommendations—that is, in almost every part of the fashion product cycle (Jain, Bruniaux, Zeng, & Bruniaux, 2017).

In this era of big data, social media has become an important source for generating innumerable points of digitized and fresh data (Blazquez & Domenech, 2017). From previewing new collections and interacting with designers and other fashion aficionados, to showing models behind the scenes at fashion events and providing the ability to make instant purchases, the fashion industry has become fast-paced and social-media-driven (Najjar, 2017). Staying connected with clients is key for today’s fashion industry. Social media makes the excitement and energy of fashion week and runway events directly accessible to fashion consumers in a globally impactful way (Bravo, 2016). Therefore, it is important for fashion brands to develop effective social media strategies to reach their target consumers and to exploit the informative value of the comments posted online by fashion consumers. In pursuit of these goals, questions emerge about how to maximize the number of social media users seeing and engaging with the posts and content of fashion companies and fashion media, how to encourage interaction with fashion consumers on social media platforms, and how to understand the full picture of what happens during fashion event cycles. The emergence of fashion informatics may contribute to answering such questions.

The researchers behind a few fashion informatics studies have tried to identify new trends in the fashion world and study the interactions between consumers and fashion brands on social media (Brambilla, Ceri, Daniel, & Donetti, 2017; D. Lee, Han, Chambourova, & Kumar, 2017). However, researchers have not sufficiently and systematically investigated the structure of fashion information and the complex interaction patterns within the fashion network. In addition, most fashion informatics research has been conducted by computer scientists, who may lack the perspective of fashion design/merchandising researchers (Reiter et al., 2016). Therefore, in this study, we applied data-mining-based SNA to demonstrate data structure and information flow during Paris Fashion Week in order to understand the dynamic fashion world on social media. Data mining, a term used in computer science, refers to the process of automated information extraction using as input a variety of complex or unstructured data sources (Feldman & Sanger, 2007). On the other hand, SNA is the process of investigating social structures through the use of networks and graph theory (Hanneman & Riddle, 2005). To the best of our knowledge, we are among the first to empirically investigate and visualize fashion network, social media data mining. Given the power of the computational approach, in this study, we aim to (a) discover potential applications for fashion informatics by crawling large-scale data sets on social media, (b) visualize the dynamic network to pinpoint key influencers and major communities during fashion event cycles, (c) investigate the structure of fashion topics on social media and the relationships among these topics, (d) explore the possibility of adapting data-mining techniques for fashion researchers and highlight the advantages of user-generated data, and (e) bridge the gap between knowledge and practice, providing insights on how to extract pertinent information from unsolicited posts on social media to support the decision-making of fashion industry businesses.

Literature Review

Social Media and Fashion Informatics

Social media use is playing a growing role in reconstructing the fashion business landscape. Many fashion companies with branded merchandise have created their own social media accounts to communicate with customers without restrictions on time, place, or medium (Kim & Ko, 2012). The rules of engagement have changed. Fashion communication has evolved from a top-down structure to a more democratic form of communicating. Although authors of previous studies have employed interviews and surveys to discover the influence of fashion blogs on fashion consumers’ motives to engage with social media platforms, the complex patterns of influence within a group and the information flow of communication models are difficult to identify (Halvorsen, Hoffmann, Coste-Manière, & Stankeviciute, 2013; Wolny & Mueller, 2013). The potential contribution of social media platforms to transform the power relationship among fashion brands, fashion media, and fashion consumers has created many exciting research opportunities that involve crawling this unstructured, user-generated data (Zafarani, Abbasi, & Liu, 2014). A combination of theoretical, computational, and statistical approaches can be used to better understand the dynamic nature of fashion trends and fashion consumers. This new interdisciplinary field—fashion informatics—may help fashion companies come to startling conclusions about their designs and revolutionize the way industrialists and brands produce apparel and accessories.

The majority of researchers doing quantitative studies related to social media in fashion have employed survey methods to collect and analyze data. Despite having many advantages, survey methods do present some limitations. First, opinions can rarely be measured continuously, which limits the ability to measure the evolution of consumers’ responses toward fashion trends or campaigns (Ceron & Negri, 2016). Second, while survey data flow from questions designed by researchers seeking to accept or not accept hypotheses, user-generated data are spontaneously created and, thus, more reflective of the thoughts and trends of influencers and consumers (Salganik, 2017). Third, standard sampling and statistical testing rely on the independence of observations, which does not hold for network data. It is challenging to explore processes of change in networks and relate structural transformations to the unanticipated consequences of individual-level decision-making (Monge & Contractor, 2003). Taking advantage of computational methodologies can help scholars overcome these limitations and adapt to the transition from the analog world to the digital world.

Researchers using powerful computational resources combined with massive social media data sets have shown, in a few studies, the role of social media analysis in fashion informatics (Ruths & Pfeffer, 2014). One stream of research is to identify new influencers and trends in fashion. D. Lee, Han, Chambourova, and Kumar (2017) collected a data set of 10,000 Twitter accounts and trained a classifier to identify whether an account was fashion related. Such a data set helps narrow the search space, aiding researchers in observing trends, microtrends, and fads. Park, Ciampaglia, and Ferrara (2016) also used social media to understand the ingredients of fashion models’ success by employing a machine-learning framework that used statistical techniques to give computer systems the ability to learn from data. They predicted the tenure of a cohort of new faces from the 2015 spring/summer season through the subsequent 2015–2016 fall/winter season. Interestingly, they confirmed that a strong social media presence may be more important to success than being under contract with a top agency or than the aesthetic standards sought after by the industry. New generations of supermodel and social media superstars, like Gigi Hadid, well and truly embody the current fashion zeitgeist.

The fascinating theme of interactions between fashion brands and consumers is the subject of another stream of literature. In one recent study, Brambilla, Ceri, Daniel, and Donetti (2017) showed how social content responds to live fashion events. Their work organized fashion brands into clusters based on similar features. By studying the timing and location patterns of social-media responses to these clusters of brands, the researchers could understand how time and space play a role in fashion activities on social media. Manikonda, Venkatesan, Kambhampati, and Li (2015) studied the top 20 fashion brands and the companies behind them in terms of the number of social media followers and compared how different companies target their current and potential customers via social media platforms. Nearly 180,000 Twitter posts and more than 34,000 Instagram posts were examined using Latent Dirichlet allocation to understand the types of topics the different companies focused on. Most of the posts on runway brands focused on the same topics across these two social media platforms, whereas posts about brands like Nike, Free People, and so on focused on different topics. Manikonda et al. (2015) also used visual analysis and textual analysis to reveal how companies focus on different topics on different social media and how certain types of visual cues associated with marketing strategies can obtain more visibility.

Despite these efforts, the authors of current fashion informatics studies have not sufficiently and systematically investigated the structure of fashion units (fashion topics, brands, consumers, and other fashion groups) or the dynamics of social content on social media. The data-mining-based SNA methods described in the next section will allow researchers and practitioners who are interested in fashion informatics to analyze interactions and information flow among fashion units and unravel complex interaction patterns within the network.

SNA in Fashion Informatics

A social network is composed of a set of socially relevant nodes (actors, values, sentiments, ideas, locations, attributes) and the relationships (links, ties, associations, affiliations, interactions, evaluations) among those entities (Cioffi-Revilla, 2014). In the case of fashion, these nodes are most commonly brands, designers, trend-forecasting organizations, consumers, and key words. However, in principle, any units that can be connected to other units can be studied as nodes. In terms of social networks on social media platforms, topics, accounts, and hashtags are often analyzed as nodes while relations among nodes include friendships, followings, trade ties, web links, citations, information flows, or any other possible connection among these identified nodes (Zafarani et al., 2014). On social media, flow-based relations reflect how fashion information is exchanged or transferred between nodes. For example, when fashion companies release new collections or advertising campaigns via their social media accounts, fashion enthusiasts who like, share, or retweet the postings show how information and influence flow through networks.

Drawing from a paradigmatic perspective of emergent social structures and the branch of mathematics called graph theory, SNA converts social networks into abstract models of points and lines (Hanneman & Riddle, 2005). That is, individuals and other social actors (nodes) are represented by the points, and their social relations (edges) are represented by the lines (Hanneman & Riddle, 2005). Several researchers in social science have addressed the links between SNA and theories of rational choice. Human beings are social animals by nature, and it is therefore natural that they use computer networks to facilitate their social connections and even widen their range of relationships. The small world theory considers societies to be close-knit structures, which are highly locally clustered with a short path length between the actors (Watts, 1999). The online social networks have been found to exhibit typical characteristics of a small world network, with short average distances between the users of less than six and a high clustering coefficient. This suggests that the information can travel faster through online social media as compared to any other traditional media. The strength of weak ties theory is used to explain the relationships and flow of information among various social networks (Granovetter, 1977). It holds that the stronger the tie between two people, the more likely that their social life will overlap and have ties with the same or a similar third person having similar opinions and ideas. A bridging tie that links a person to people who are not connected to his or her other friends is the source of novel ideas and information, as people in one network are different from those in another and may have a different opinion or new ideas. These bridging ties are usually weak ties between the two members of different communities but hold greater strength in the sense that they are the best potential source of novel information. With regard to measuring important nodes in networks, several measures are introduced in Figure 1—indegree, outdegree, betweenness, and closeness—that have been used to measure the importance of a node in the network from various perspectives (Adamic, 2014).

Figure 1.

From left to right, measures in social network: indegree, outdegree, betweenness, and closeness. Adapted from Adamic (2014).

Degree centrality is the number of nodes that are directly linked with a node (Freeman, 1978). In a directed network, indegree is the number of edges that are directly linked to a node, while outdegree is the number of edges that are directly linked from a node. If a node has direct links to other nodes, the node occupies the central role and has a big impact in the network, such as node X in the first two panels of Figure 1. A second measurement is betweenness centrality. If a node is passed through by the shortest path of two other nodes, the node occupies an important location where it controls the communication of the other two nodes. The more such locations a node occupies, the greater its betweenness centrality. The betweenness centrality of node i is defined as: $C_{B} (i) = \sum_{j \neq k} g_{j k} (i) / g_{j k}$ where g_jk(i) = the number of shortest paths connecting node j and node k passing through node i, and g_jk = total number of shortest paths. If a node links to many other nodes through shortest paths, it has high closeness centrality, which means the node is located in the central part of the network and is less controlled by others (Freeman, 1978). Closeness centrality is the inverse of the sum of the distances between a node and all other nodes in the graph. It is defined as $C_{C} (i) = {[\sum_{j \neq i} d (i, j)]}^{- 1}$ where d(i,j) = the length of the shortest path between node i and node j. Lastly, K-core is a network metric that can help to identify strongly connected nodes when the network is too large to consider all nodes. A K-core is the largest group of nodes in the network where all nodes are linked with at least K nodes in the group. It can be obtained by recursively deleting nodes with fewer than K edges (Freeman, 1978).

Analyzing networks has many advantages, such as (a) moving beyond individual perception and thinking of individuals as embedded in groups to “varying degrees and thus differentially subject to the opportunities, constraints and influences created by group membership” (Marin & Wellman, 2011, p. 14), (b) defining the connectivity of groups and whether they are more cohesive or more permeable, and (c) exploring vague structures in social relations that are not clearly identifiable groups, such as weak ties or acquaintances (Freeman, 1978). Having access to these substantial advantages, SNA has been widely used to find the overall characteristics, internal construction, and core-periphery structure of networks (Himelboim, Smith, Rainie, Shneiderman, & Espina, 2017; K. Lee, Jung, & Song, 2016). It has proved useful in quantifying the associations between words, thus showing good performance in reflecting the semantic structures of topics from specific domains such as social media (Himelboim et al., 2017). In order to extract fashion-related data on social media, the computer-based approach called data mining becomes a powerful method in the analysis of networks (Cioffi-Revilla, 2014). Data mining can be used to extract social media information pertinent to nodes and edges that constitute networks present in source data. In other words, social media data can be mined automatically to extract various kinds of societal network structures of interest (Cioffi-Revilla, 2014).

A few scholars have explored the potential application of SNA in the fashion field with a traditional data-collecting approach. Reiter, Zhao, Ciampaglia, and Ferrara (2016) discovered fashion knockoff networks by using network science techniques. A fashion knockoff network was visualized to understand knockoff victims and offenders. Song, Hwang, Kim, and Kwak (2013) investigated the acceptance of word-of-mouth information between consumers in the Internet fashion community. SNA helped reveal the features of the network and its spread of patterns involving fashion information shared between community members. Despite this important work that provided a novel examination of information flow’s network structure in the fashion community, a traditional survey method was employed to collect the data. The authors did not capture the entire available data set on social media by utilizing the computational approach, leaving a gap in our understanding of how data-mining techniques can be applied in the fashion world to discover patterns in complex networks. Therefore, it is critical to scale up the implementation of big data and the computational approach in SNA in the fashion field to demonstrate the novel link between fashion and informatics. In this study, data-mining-based SNA was used to investigate online discussions on fashion-related topics to exhibit the implementation of SNA in the fashion field.

Application in Fashion: Paris Fashion Week

Context and Background

We investigated one case study related to the 2017 Paris Haute Couture Fashion Week (July 2–July 6) to demonstrate how to understand fashion industry practices by adopting data-mining-based SNA. Social media activities related to key fashion events, such as Fashion Week, have drawn much attention from practitioners and scholars. Fashion Week is a prominent and exclusive fashion event that sets new trends and gives the world a peek at the latest in high-end couture (Bravo, 2016). The Haute Couture show is an exclusive event and only held during Paris Fashion Week (Chevalier & Mazzalovo, 2008). Fashion-related companies host a variety of social media activities before, during, and after Fashion Week to increase customer engagement. Through social media, many genius designs, dreamy creations, beautiful models, A-list celebrities, and shocking backstage stories become the center of attention and earn buzz and mentions all around the world.

Twitter was chosen as the social media platform for this study. Twitter is particularly well suited as a source of real-time event content. Its lightweight interface, suitable to mobile devices and accessible via texting, means that it is often the platform of choice when on the move, in low-bandwidth environments or at high-tension events (Bruns, Highfield, & Burgess, 2013). The hashtag (#) symbol is used on Twitter to enable the manual or automatic collation of all tweets containing the same hashtag. Analyzing the hashtag community is one of the most common ways to research the flow of information on Twitter (Bruns et al., 2013). Coverage of Fashion Week on Twitter provides examples of the formation of fashion topics through shared hashtags. By including fashion-related hashtags in their tweets, social media users are connecting their comments to a wider discussion. We selected three hashtags related to Paris Fashion Week: #pfw (Paris Fashion Week), #Hautecouture (Haute Couture), and #Chanelhautecouture (Chanel Haute Couture). The hashtag “pfw” is the official tag for Paris Fashion Week and is managed by the Twitter account of the Fédération de la Haute Couture et de la Mode (@fhcmode). The Haute Couture hashtag is widely used by fashion companies during Fashion Week. Chanel was chosen because it has the most followers on Twitter among all brands attending the Paris Haute Couture Fashion Week. The hashtag #Chanelhautecouture was the tag used by the official Chanel account for promoting the Chanel Haute Couture fashion show on Twitter.

Method

To identify topics and quantify the frequency and other aspects of the online discussion among Twitter users over time, we created a series of networks of hashtags connected through user tweets. In these networks, nodes are hashtags created by users when they publish tweets on certain events. Through the Twitter Application Programming Interface (API), tweets containing the three chosen hashtags were crawled the week before (6/25-7/1), during (7/2-7/6), and the week after (7/7-7/13) Paris Fashion Week to monitor the mobilization of topics related to this event. We then extracted all hashtags, which are naturally good representations of topics being discussed, from the tweets’ data. These became the initial nodes for the networks. In addition, noise was carefully reviewed and removed. An overview of the crawled tweets’ data, number of nodes, and number of edges in the networks is shown in Table 1.

Table 1.

Overview of the Crawled Tweets’ Data, Number of Nodes, and Number of Edges in the Networks.

Hashtag	Before PFW			During PFW			After PFW
Hashtag	No.	Nodes	Edges	No.	Nodes	Edges	No.	Nodes	Edges
#pfw	1,859	90	370	3,902	803	3,262	1,249	176	662
#Hautecouture	1,297	896	3,950	3,562	1,864	8,697	2,450	1,533	6,826
#Chanelhautecouture	7	2	1	646	210	521	157	121	280

Note. PFW = Paris Fashion Week.

To create edges between nodes, we looked to the co-occurrence of hashtags in the same tweet, connecting two nodes if they occurred in the same tweet. The strength of the edge grows proportionally to the number of tweets that contained both nodes linked by the edge. The network visualization software Gephi was applied to analyze and visualize these topic networks generated by the Twitter hashtags. As there can be hundreds of nodes and thousands of edges in the networks, we applied filters using some conditions (e.g., limiting node degree, K-core, or tag frequency) to present the core structure of topic discussion networks. Then, Gephi’s algorithms were used to roughly categorize the nodes into clusters, where hashtags in the same cluster are viewed to discuss related topics. To show the division of different subtopics in a topic network and for a better visualization effect, the layout algorithm Force Atlas was applied and the nodes and edges were colored according to the groups that the nodes belong to. As the closeness centrality metric does not differentiate nodes (hashtags) in the networks of this study, we only report the values of degree and betweenness centrality in the Results section.

Results and Discussion

Case 1: Paris Fashion Week (#pfw and #ParisFashionWeek)

Data representing #pfw network visualization before, during, and after Paris Fashion Week can be found in Figure 2. We first looked at the most influential nodes within three networks of #pfw. The node with the higher betweenness centrality is more influential in the network because it functions as a junction for communication within the network (Freeman, 1978). Before Paris Fashion Week, #pfw, #fashion, #mensfashion, #menswear, and #ss18 were the most important junctions for meaning circulation within the network. The Paris Fashion Week hashtag (#pfw) was the central term and was adjacent to the most hashtags in the network. According to the official fashion week schedule, Paris Men’s Fashion Week was held before Haute Couture Week. Attention of the fashion world gradually shifted from Paris Men’s Fashion Week to Haute Couture Week. During the Paris Fashion Week, the convergence of topics can be observed. The largest component of the network was widely interconnected; therefore, the topic being discussed was a major topic in this network. The hashtags paris, hautecouture, fashion, couture, fashionweek, hollywood, and model were the most influential junctions within the network. After Paris Fashion Week, topics were moving away from #hautecouture and #paris. However, #fashionweek was still one of the most frequently employed hashtags. Following the fashion week schedule, hashtags for the London, Milan, and New York fashion weeks started to get attention from fashion followers, and these became the most influential topics affecting the fashion industry.

Figure 2.

From top to bottom, visualized #pfw network before, during, and after Paris Fashion Week.

In addition, distinct hashtag communities are present within the network. Before Paris Fashion Week, the top three contextual clusters within this network were (a) #bestdressedmen, #bestdressedman, #bestdressed, #alexbadia, #streetstyle, #theimpression; (b) #top, #lighting, #new, #luxury, #brass, #design; and (c) #sffw, #lafw, #nyfw, #lfw, #fwla, #fwny, #siliconbeach_la. The largest community, in red in Figure 2, included followers interested in voting for the “best dressed” people of Fashion Week. The second largest community, shown in blue, focused on the designs of the runway decorations. Notably, the third largest community, in green, included those who linked their attention to other fashion weeks, such as New York Fashion Week (#nyfw) and London Fashion Week (#lfw). Attention was drawn to California as the next fashion center, with Los Angeles Fashion Week and San Francisco Fashion Week catching people’s attention.

During Paris Fashion Week, frequently co-occurring hashtags were #couture, #hautecouture, #paris, and #fw17, which indicated the theme of the fashion week. Some attention-grabbing clusters rose to the top. The red color represents important components of fashion week including #paris, #fashion, #model, #style, #runway, #backstage, #fashionshow, #fashionblog, and many luxury brand names. That is, when fashion followers talk about “Paris” and “fashion” on social media, they will most likely pay attention to fashion models and various fashion shows hosted by luxury brands. The second important topic, in blue, represents the interconnected cluster of #haute #collection, #lfw, and #mfw. This demonstrates that haute couture and luxury were associated with the country of origin’s culture and were mainly a European phenomenon. Most invited and foreign members of the Chambre Syndicale de la Couture Parisienne are from Italy and the United Kingdom (Fhcm, 2017). In addition, #ootd (outfit of the day), #facebook, #followme, and #love, in purple, formed another cluster and indicated that people love showing others what they have worn across different social media platforms. Several semantic networks emerged, such as #hautecouture and #couture, #fw17 and #aw17.

After Paris Fashion Week, the top contextual community, in yellow, included those discussing hand embroidery, which is an essential technique for haute couture and is linked to several fashion brands (#tomford, #etro, #ralphlauren) and past fashion weeks (#ss13, #fw13, #fw14, #fw15). In particular, #embroidery had a high degree (22), but lower betweenness centrality (42.060) than other words. This indicated the hashtag embroidery was an important local hub that binds together a cluster of terms that form a specific context, but it is not as central as other important terms to the network as a whole. The top two contextual communities, in blue, showed that the United States, specifically California and New York, attracted much attention after Paris Fashion Week. One possible reason is the impact Hollywood celebrities have on the fashion industry. The third community, in red, contained followers discussing various activities around the time of fashion week, such as #hiphop, #stylist, #rocker, and #fashionbloggers. Interestingly, one cluster, in green, that emerged in the network included people interested in specific styles and design elements. Fashion trends now started involving and affecting fashion followers in a trickle-down process. Paint, Egyptian, stripes, and trenchcoat were trending hashtags. Table 2 provides the top 10 nodes with degree information, and the top 10 nodes in the network and their betweenness centrality.

Table 2.

Case 1: Twitter Top 10 Nodes With Degree Information and Top 10 Nodes in the Network and Their Betweenness Centrality.

Before		During		After
Label	Degree	Label	Degree	Label	Degree
pfw	97	paris	255	paris	53
fashion	44	hautecouture	195	lfw	47
menswear	29	fashion	185	fashion	45
mensfashion	25	couture	144	fashionweek	41
ss18	25	fashionweek	106	mfw	39
runway	15	fw17	93	hautecouture	37
menstyle	15	style	87	nyfw	35
paris	15	ss18	80	style	27
style	15	aw17	77	couture	24
sffw	14	model	63	collection	23
Label	Betweenness	Label	Betweenness	Label	Betweenness
pfw	3,526	paris	13,332	paris	516
fashion	321	hautecouture	7,217	fashion	386
menswear	72	fashion	6,873	lfw	269
ss18	54	couture	3,595	fashionweek	264
mensfashion	45	fashionweek	1,780	hautecouture	230
hautecouture	31	hollywood	1,570	mfw	183
paris	24	fw17	1,315	nyfw	123
runway	20	ss18	1,308	style	96
couture	7	style	1,214	ss18	82
fashionblogger	4	aw17	1,045	couture	70

Case 2: Haute Couture (#HauteCouture)

The second hashtag studied, HauteCouture, was examined to investigate information flow about specific product categories on social media. Data representing #HauteCouture network visualization before, during, and after Paris Fashion Week are shown in Figure 3. Before Paris Fashion Week, people discussed fashion, moda (French: model), and luxury. Fashion models showed their influential power in the network. Fashion followers were looking forward to bridal and dress inspiration. Some key words related to couture emerged, such as #handmade, #elegant, #art, #fashionista, and #redcarpet. Interconnected hashtag clusters, in blue, were specifically focused on the models of two brands: Dior and Gucci. During Paris Fashion Week, the largest component of the network is widely interconnected. The Haute Couture hashtag became the focus of the fashion world. The companies behind some luxury brands also demonstrated their influential power in the haute couture world, such as Chanel, Dior, and Elie Saab. Both French groups (#moda and #altacostura) and English groups emerged in the network. After Paris Fashion Week, topics remained connected with #HauteCouture, which was different from the case of #pfw. The hashtags fashion, paris, and couture remained very influential in the network. One community, in green, was focused on #fashion, #style, #fashionista, #elegant, #chic, and #beautiful. In particular, #fashion functions as a junction between #paris and #style. It indicated the city image of Paris was closely related to fashion. The top 10 nodes with degree information, and the top 10 nodes in the network and their betweenness centrality, are shown in Table 3.

Figure 3.

From top to bottom, visualized #Hautecouture network before, during, and after Paris Fashion Week.

Table 3.

Case 2: Twitter Top 10 Nodes With Degree Information and Top 10 Nodes in the Network and Their Betweenness Centrality.

Before		During		After
Label	Degree	Label	Degree	Label	Degree
fashion	58	paris	255	fashion	443
couture	40	fashion	195	paris	280
moda	38	pfw	185	couture	264
style	34	couture	144	pfw	243
luxury	31	fashionweek	106	style	178
ss17	27	fw17	93	fashionweek	143
bridal	27	style	87	fashionshow	102
dress	26	chanel	80	moda	97
paris	26	mode	77	mode	96
runway	25	moda	63	fashionista	93
Label	Betweenness	Label	Betweenness	Label	Betweenness
fashion	19,406	paris	71,914	fashion	41,496
couture	4,366	fashion	50,958	paris	15,137
love	2,558	pfw	49,855	couture	13,313
details	2,295	couture	23,251	pfw	11,116
style	2,143	fashionweek	7,654	style	4,815
paris	2,047	fw17	7,374	fashionweek	3,255
moda	1,611	style	5,353	atelier	1,621
dress	1,487	chanel	4,187	designer	1,552
Art	866	mode	2,867	fashionshow	1,365
bridal	640	collection	2,303	moda	1,331

Case 3: Chanel Haute Couture (#Chanelhautecouture)

In order to investigate how fashion companies leverage Twitter to amplify a big event, #Chanelhautecouture was chosen as the third case in this study. Chanel was shown to be the most influential luxury brand in Case 2, according to the betweenness centrality. Chanel distributed the designated branded hashtag #Chanelhautecouture and #Chaneltower on Twitter on July 4, so the network of the week before Fashion Week is left out. Figure 4 indicates the #Chanelhautecouture network during and after Paris Fashion Week. During Paris Fashion Week, the Chanel account had a spike of likes and comments, indicating that the content it was publishing was highly compelling to its audience. Frequently co-occurring hashtags included #Chaneltower and #KristenStewart. As a high-profile celebrity and official spokesperson for the Gabrielle bag campaign, Kristen Stewart is an important influencer and closely correlated with Chanel. Compared to Kristen Stewart, Cara Delevingne, who is also a spokesperson for Chanel, showed less influential power in the network. Interestingly, Katy Perry showed a large influence in the green cluster. Perry is not the official face of Chanel and only connected with Chanel through Paris Fashion Week. After Paris Fashion Week, more focus moved to styles and inspirations. Celebrities Kristen Stewart and Julianne Moore remained big influencers in the network. More than eight clusters were clearly divided and emerged in the network. For example, the green community was composed of those who paid attention to #artisan, #classics, and #elegance. The yellow cluster was focused on #style, #UK, #Europe, and #london. Members of the purple community discussed the inspiration for work suits. Table 4 indicates the top 10 nodes with degree information and the top 10 nodes in the network and their betweenness centrality.

Figure 4.

From top to bottom, visualized #Chanelhautecouture network during and after Paris Fashion Week. Before is left out.

Table 4.

Case 3: Twitter Top 10 Nodes With Degree Information and Top 10 Nodes in the Network and Their Betweenness Centrality.

During		After
Label	Degree	Label	Degree
chaneltower	33	chanel	41
chanel	30	fashion	35
fashion	25	chaneltower	28
paris	20	kristenstewart	20
fashionweek	16	style	13
pfw	16	hautecouture	12
kristenstewart	11	artisan	9
katyperry	11	tweed	9
style	11	story	9
hautecouture	11	black	9
Label	Betweenness	Label	Betweenness
chaneltower	158	chanel	321
chanel	113	fashion	217
fashion	54	chaneltower	146
paris	26	kristenstewart	45
fashionweek	12	style	16
karllagerfeld	8	hautecouture	12
hautecouture	6	show	2
kristenstewart	6	juliannemoore	1
pfw	5	repost	1
pariscoutureweek	4	no5	0.25

Conclusions and Implications

Data-mining-based SNA is a promising area of fashion informatics for investigating relations and information flow among fashion units. By adopting this pragmatic approach, we provide dynamic network visualizations of the case of Paris Fashion Week. It contributes to the academic and professional discussion about how a large-scale social media data set is able to reshape the fashion industry. The pfw hashtag (Paris Fashion Week) was analyzed first to show how the focus of social media users was mobilized along with the Fashion Week time line. Olivia Palermo was the most influential fashion icon during Paris Fashion Week among the many celebrities sitting front row. In addition, the case of #pfw suggests that fashion companies may consider tagging high-profile celebrities to increase the number of consumers viewing and engaging with posts and contents on social media. Also, while preserving the traditions of haute couture, such as craftsmanship and heritage, incorporating high technologies into today’s fashion design could catch the attention of modern consumers. Second, the Haute Couture hashtag was examined. The main influential luxury brands and local hubs were identified in the network. Interestingly, key characteristics of haute couture were well summarized in the network by social media users. In addition, network topics revealed that social media users often obtain inspiration about wedding dresses and evening dresses at the fashion week event. Even though people in other fashion capitals nowadays are tapping into the couture world, Paris remains the most powerful influencer in the fashion industry. Third, the most influential luxury brand, Chanel, was discussed to demonstrate how fashion companies take advantage of social media to amplify a big event. The pattern showed that #Chaneltower was a very successful theme and attracted much attention on social media. High-profile celebrities and fashion bloggers helped promote fashion brands on social media. At the same time, editors and fashion magazines, the traditional gatekeepers of the fashion industry, did not show a great deal of influence in the network. The influence of celebrities was confirmed and ranked. As influencer business is important in fashion, fashion companies should wisely choose celebrities and formulate their most effective social media campaigns to reach target audiences.

With this study, we are among the few researchers who have provided an in-depth discussion and demonstration of big data implementation in fashion through data-mining-based SNA. As a novel approach in fashion, this type of analysis is useful in detecting and predicting fashion events and trends. Visualized social networks may reduce complexity and enable researchers to easily point out key participants and clusters within the networks. This study’s design and findings allow us to show the possibility of using social media data for fashion research. Accessing social media data through API can offer fashion researchers opportunities to acquire not just public, spontaneous data, but large volumes of data, which is significant in terms of speed, reliability, and cost. This stream of information could be integrated with traditional data sources and collection methods for fashion researchers. In addition, social network theories were revisited, and for the first time applied to explain a complex relationship in the context of fashion. This empirical work extended social network theories to investigate how information flows within and between the fashion units. The present work also allows researchers to theorize about fashion communication, focusing on consumers’ engagement in electronic word of mouth. Using data-mining techniques, SNA can be used by fashion companies to identify a small number of key members of a network to promote new campaigns and adopt new products. Through key influencers in the network, fashion companies could maximize the number of social media users seeing and engaging with posts. Fashion companies could also identify important nodes in a network to discover how communities are connected to strategically engage fashion consumers. Lastly, the rise of fashion informatics can lead fashion educators to think of how to prepare fashion students for the world of big data. Curriculum change is necessary to teach students the advanced quantitative skills they will need to work with massive data sets.

Limitations and Future Research

There are several limitations to this study, which may lead to future research opportunities. First, only hashtag networks were observed in this study. Networks using following, @replies, or retweets could be interesting for researchers to explore. Such a conversational context would be valuable to extract more sophisticated social media data. Second, both French groups and English groups emerged in the network. Future researchers may want to consider how social media users with different geographical locations discuss the same fashion event. Third, social media users may not be representative of the whole population of fashion consumers. Future research may be needed to clearly identify the social demographic traits of social media users and recommend data analysis strategies for high-end, fast fashion, or sustainable fashion brands.

Footnotes

Acknowledgments

We thank Jefferson Henrique for providing Python codes for crawling historical tweets, which are available here

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Li Zhao

Chao Min

References

Adamic

(2014). SNA 3A centrality [PDF document]. Retrieved from https://github.com/ladamalina/coursera-sna/tree/master/Week%203.%20Centrality

Blazquez

Domenech

(2017). Big data sources and methods for social and economic analyses. Technological Forecasting and Social Change, 130, 99–113. doi:10.1016/j.techfore.2017.07.027

Brambilla

Ceri

Daniel

Donetti

(2017, April). Spatial analysis of social media response to live events. Paper presented at the 26th International Conference on World Wide Web Companion, Perth, Australia.

Bravo

(2016, February). How social media has changed the fashion week game. Retrieved from http://www.sfchronicle.com/style/article/How-social-media-has-changed-the-fashion-week-game-6855195.php

Bruns

Highfield

Burgess

(2013). The Arab spring and social media audiences: English and Arabic Twitter users and their networks. American Behavioral Scientist, 57, 871–898.

Ceron

Negri

(2016). The “social side” of public policy: Monitoring online public opinion and its mobilization during the policy cycle. Policy & Internet, 8, 131–147.

Chevalier

Mazzalovo

(2008). Luxury brand management: A world of privilege. Hoboken, NJ: John Wiley.

Cioffi-Revilla

(2014). Introduction to computational social science. New York, NY: Springer.

Feldman

Sanger

(2007). The text mining handbook: Advanced approaches in analyzing unstructured data. Cambridge, England: Cambridge University Press.

10.

Fhcm. (2017, June). Paris haute couture fashion week. Retrieved from https://fhcm.paris/en/

11.

Freeman

L. C.

(1978). Centrality in social networks conceptual clarification. Social Networks, 1, 215–239.

12.

Granovetter

M. S.

(1977). The strength of weak ties. Social Networks, 78, 347–367.

13.

Halvorsen

Hoffmann

Coste-Manière

Stankeviciute

(2013). Can fashion blogs function as a marketing tool to influence consumer behavior? Evidence from Norway. Journal of Global Fashion Marketing, 4, 211–224.

14.

Hanneman

Riddle

(2005). Introduction to social network methods. Riverside: University of California, Riverside.

15.

Himelboim

Smith

M. A.

Rainie

Shneiderman

Espina

(2017). Classifying Twitter topic-networks using social network analysis. Social Media + Society, 3. doi:10.1177/2056305117691545

16.

Jain

Bruniaux

Zeng

Bruniaux

(2017, October). Big data in fashion industry. Paper presented at IOP conference series: Materials science and engineering, Busan, Korea.

17.

Kim

A. J.

(2012). Do social media marketing activities enhance customer equity? An empirical study of luxury fashion brand. Journal of Business Research, 65, 1480–1486.

18.

Kindred

Steele

(2015) Fashioning data: A 2015 update. Data innovations from the fashion industry. Retrieved from http://www.oreilly.com/data/free/fashioning-data.csp

19.

Lee

Han

Chambourova

Kumar

(2017, August). Identifying fashion accounts in social networks. Paper presented at conference on knowledge discovery and data mining, Halifax, Canada.

20.

Lee

Jung

Song

(2016). Subject–method topic network analysis in communication studies. Scientometrics, 109, 1761–1787.

21.

Lin

Zhou

(2015, January). Text-generated fashion influence model: An empirical study on style.com. Paper presented at 48th Hawaii International Conference on IEEE, Honolulu, HI.

22.

Manikonda

Venkatesan

Kambhampati

(2015, December). Trending chic: Analyzing the influence of social media on fashion brands. Retrieved from https://arxiv.org/pdf/1512.01174.pdf

23.

Marin

Wellman

(2011). Social network analysis: An introduction. In Scott

Carrington

P. J.

(Eds.), The SAGE handbook of social network analysis (pp. 11–25). London, England: Sage.

24.

Monge

P. R.

Contractor

N. S.

(2003). Theories of communication networks. Oxford, England: Oxford University Press.

25.

Najjar

(2017). Is social media making fashion week obsolete? Retrieved from https://www.popsugar.com/fashion/Social-Media-Effect-Fashion-Week-43165848

26.

Park

Ciampaglia

G. L.

Ferrara

(2016, August). Style in the age of Instagram: Predicting success within the fashion industry using social media. Paper presented at Conference on Knowledge Discovery and Data Mining, San Francisco, CA.

27.

Reiter

Zhao

Ciampaglia

Ferrara

(2016, August). Big data in fashion: A study of fashion knockoff networks. Paper presented at Conference on Knowledge Discovery and Data Mining, San Francisco, CA.

28.

Ruths

Pfeffer

(2014). Social media for large studies of behavior. Science, 346, 1063–1064.

29.

Salganik

(2017). Bit by bit: Social research in the digital age (Open Review Ed.). Princeton, NJ: Princeton University Press.

30.

Song

Hwang

Kim

Kwak

(2013). The effects of social network properties on the acceleration of fashion information on the web. Multimedia Tools and Applications, 64, 455–474.

31.

Watts

D. J.

(1999). Networks, dynamics, and the small-world phenomenon. American Journal of Sociology, 105, 493–527.

32.

Wolny

Mueller

(2013). Analysis of fashion consumers’ motives to engage in electronic word-of-mouth communication through social media platforms. Journal of Marketing Management, 29, 562–583.

33.

Zafarani

Abbasi

M. A.

Liu

(2014). Social media mining: An introduction. Cambridge, England: Cambridge University Press.