New Hidden Persuaders: An Investigation of Anchoring Effects of Recommender Systems on Consumer Choice
Abstract Today, websites try to suggest us everything, ranging from our next
sweatshirts to our next holiday destinations and even our next friends. Based on past
behavior or preference statements, recommender systems implemented on marketers’ webpages generate personalized predictions about the products, countries, or
people we presumably like and preselect a small subset comprising only potentially
relevant choice options from the large number of available alternatives (e.g., Ansari
et al. 2000; Bodapati 2008; Resnick and Varian 1997). This preselection is intended
to reduce decision efforts and uncertainty (e.g., Ansari et al. 2000; Herlocker et al.
2000; Tam and Ho 2005) and, thereby, to increase sales, customer satisfaction, as
well as loyalty (e.g., Fleder and Hosanagar 2009; Jannach and Hegelich 2009;
Senecal and Nantel 2004; Pathak et al. 2010; Pu et al. 2011). While extant studies
are typically based on the assumption that a recommender system’s effectiveness
depends on its ability to identify consumers’ preferences and, therefore, is measurable by the number of individuals who follow the generated suggestions, the present
study, in contrast, focuses on the impact of recommendations on the construction of
preferences. In line with the large body of research on behavioral decision-making,
preference structures are often not well defined or stable (e.g., Bettman et al. 1998;
Slovic 1995) such that preferences are frequently formed during decision-making
rather than before and, thus, might be sensitive to recommended items.
The results of two experimental studies, conducted in different contexts (i.e.,
backpacks and hotels), reveal that recommendations provided by recommender systems cause an anchoring effect on consumer choices which is reflected in biased
decisions toward (even randomly) generated suggestions. More precisely, even if
recommendations do not match preferences, people still seem to adapt their decisions to the characteristics of recommended items. Thus, this research contributes to
research on recommender systems in the marketing discipline (e.g., Ansari et al.
2000; Bodapati 2008; Senecal and Nantel 2004.
Detecting Conflict on Social Media
The majority of companies have now adopted social media. The initial euphoria
framing social media as a golden opportunity to build harmonious communities of
consumers has faded. Marketers now take a more balanced view, giving attention to
disharmonious interactions between consumers on social media. Specifically, social
media conflict attracts increasing interest from academics and practitioners because
of its disruptive nature and potential destructiveness. It has a strong influence on
value formation, whether by creating or destroying value. However, existing monitoring tools fail to adequately detect social media conflict, leaving marketers unable
to manage conflict effectively. This research develops an instrument to automatically
detect social media conflict. In this paper, social media conflict is first conceptualized
before presenting the methodology and findings and discussing implications.
Because social media conversations are rooted in feelings of kinship and togetherness, the majority of research on value formation on social media has focused on how
consumers derive value from harmonious interactions with fellow consumers who
share similar interests (Kozinets 1999). However, a growing body of articles has
highlighted that disharmonious interactions where the goals of consumers are misaligned are pervasive, influencing value formation. Researchers have thus investigated phenomena like negative online word of mouth (Babic et al. 2015),
consumer-to-consumer tensions (Chalmers-Thomas et al. 2013), consumer-to-business tensions (Kozinets et al. 2010) or social problems and social control (Sibai et al.
2015). Most recently, conflict has been highlighted as an important phenomenon
(Husemann et al. 2015). From arguments to frictions, discords, dispute, controversies and quarrels, between 15 and 40% of conversations on social media are conflictual (Johnson et al. 2008; Mishne 2007). As an extreme event, social media conflict
receives much attention, is well memorized and carries high weight when forming
judgements about an interaction or the community at large (Lea et al. 1992). Conflict
therefore heavily influences value formation on social media, both destroying and
creating value (Husemann et al. 2015). In addition, when conflict takes the form
cyber harassment, it can have dramatic consequences for individuals, and it is social
media marketers’ responsibility to intervene if this happens on their platform.
Conflict is generally defined as a series of interactions where two or more parties
manifest the belief that they have incompatible interests (Kriesberg 2007). Social
media conflict differs from conflicts unfolding in other contexts because it is technology mediated and public. Technology mediation implies that conflict behaviours
cannot directly impact online users’ physical health or economic resources.
Aggressions rather aim to harm the other party’s “face”, the public self-image that
every individual wants to claim for themselves (Brown and Levinson 1987). Online
aggressions are therefore face-threatening acts (FTAs) or impolite behaviours, that
is, acts hurting the addressee’s wish to be accepted or liked by others (Brown and
Levinson 1987; Perelmutter 2013). Because interactions in social media are public,
conflicts typically involve a large number of participants (Gebauer et al. 2013;
Lorenzo-Dus et al. 2011). All members cannot talk at the same time and equally
drive the conversation. A few posters actively drive the conversation, while the
majority take a backseat, watching and commenting on the conversation. The roles
of participants in social media conflict are thus divided between that of performer
and audience (O’Sullivan and Flanagin 2003; Perelmutter 2013), so that social
media conflicts are performances (Sibai 2014). Social media conflicts are thus
defined as performances where consumers, administrators, community owners or
companies oppose one another by engaging in face-threatening acts in front of an
audience on a social media platform.
Methodology
The context of the research is a UK-based forum for fans of electronic dance music
(EDM). Over the past 14 years, approximately 20,000 people joined the forum, posting over seven million posts. This context was selected because it contains numerous, lengthy and very diverse conflicts, therefore allowing to observe a variety of
social media conflicts. The research was conducted in two phases. A netnography
(Kozinets 2010) of the community was first conducted to develop a theory of social
media conflict and its consequences for value formation. Then, forum comments
were mined to determine the linguistic characteristics of social media conflict.
The netnography combined seven in-depth interviews (12 h of discussion, 240
pages of transcript) and 100 archived discussion threads, 32 of them being discussions about conflict (2543 posts) and 68 of them being examples of conflict (11,474
posts). Hermeneutic interpretations led to the definition of two types of conflicts:
personal and played conflicts. In personal conflicts, participants are not aware that
the conflict is performed in front of audience (e.g. flames or cyber harassment),
while in played conflicts, they are aware of it (e.g. banter conflict, cathartic conflict,
redressment rituals). Personal conflicts generate negative experiences, reducing
user engagement and community cohesion, overall destroying value, while played
conflicts have opposite consequences. Within each type of conflict, participants can
take on the role of a party who engages in attack and defence behaviours or the role
of an onlooker who engages in judgement, mediation or side comments.
In a second phase, the linguistic characteristics of social media conflict were
investigated. Twelve threads (878 posts) which are examples of conflict were
selected by convenience sampling. Each post was associated with three different
codes respectively qualifying posters’ interpretation of the conflict type, their role in
the conflict and their type of conflict behaviour. Coding criteria were defined iteratively until a coding book with 100 criteria emerged. Criteria focused on symbolic
(meaningful) rather than manifest (linguistic) features of the posts to capture
posters’ communicative intentions. The coding book was judged satisfactory when
the codes were exhaustive and mutually exclusive.
No comments:
Post a Comment