smsociety15 has ended
Back To Schedule
Wednesday, July 29 • 10:46 - 12:15
"Societal and temporal differences in herding behaviour revealed on customer reviews"

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

Authors: Dongho Choi and Chirag Shah

A variety of social media helps individuals now produce and share their information and knowledge over the online platform much easily. While, as a result, the influence of “word of mouth” is increasing faster, the shared information is not perceived equally by other people. For example, ‘most popular’ news that is highly ranked by previous viewers’ reactions and preference makes influences to following readers a lot more than rarely written news. Previous research indicated that disclosing prior collective opinions affect individual’s decision-making and as well as their perceptions of information, which is called herding effect. 

In the meantime, social media are promoting people to create high quality and quantity information for the stock of knowledge in their communities, such as Elite reviewers of Yelp.com. The assumption in this kind of user classification is that so-called Elite reviewers make substantially greater influences on other users’ information behaviours. 

We aim to to see if there exists the difference in herding effect by different groups of people, that and if so, to explore how does the herding behaviour look like in different communities and/or societies. We have three research questions as follows: (1) To what extent, if any, do users in different societies show different patterns of herding effect?; (2) To what extent, if any, do different product/service categories affect the herding effect?; and (3) In terms of temporal dynamics, has herding effect been changing over time? 

Yelp’s recently publicized data set, which includes more than 1.6 million reviews about businesses in 10 cities across 4 countries, is used to observe the rating histories of particular businesses. In addition to the rating data, other factors of reviews written by different groups of people will be extracted. More specifically, the factors extracted from reviews of elite and non-elite reviewers are trained and tested to predict current rating based on the previous history. Various classification model are used to compare the performance of our model, during which different weights are assigned to two review sets to find out the optimal weight distribution over different locations and time for consideration. 

From the preliminary study, we have found that one of the classifiers, nearest neighbor classifier show the best prediction results when we weigh more influences on elites’ reviews than non-elites’ reviews (i.e., 0.7 for elites’ reviews vs. 0.3 for non-elites’), for the sample reviews from particular businesses. We are analyzing data from popular businesses in selected categories, and in different cities in the dataset. We expect to see another societal and temporal differences from this. 

Future Work: 
We will consider more factors, such as textual data and social network data, as well as time and location data into the model in order to understand how the influential people, or their influences work differently in different societies. Also, we will validate the findings with other customer reviews data sets or different types of social media, such as Twitter, to see if the locational and temporal differences still exist. 

Wednesday July 29, 2015 10:46 - 12:15 EDT
(7th Floor) Room TRS1-129 (Ted Rogers School of Management) 55 Dundas Street West, Toronto, ON M5G 2C3

Attendees (0)