Analysis of business ranking for a connected group of Yelp users by aggregating preference pairs
Year: 2015
Type: article
Abstract: The aim of this thesis is to analyze algorithms that efficiently rank businesses (restaurants, bars, theatres, cafes etc.) of a social network.In this thesis, we focus on the analysis of algorithms applied to a data set obtained from aggregated pair preferences.The algorithms we analyze include Zermelo's probability model, PageRank, and In-degree counting.We demonstrate significant differences in the results and performance of these diverse methods.Our results will ultimately allow us to provide more relevant and better recommendations to a social network user.The main idea behind the work is perform preference pair aggregation to rank nodes of a business graph.A preference pair {A, B} for a user U suggests that -the user U prefers (better likes) business A to B. We use implicit feedback from the dataset to decide a user's preference.We try to aggregate such preference pairs for all businesses the user U and his friends F 1 , F 2 , F 3 … have reviewed to bring out meaningful recommendations for the user U. Once we have a large set of preferences pairs for a particular user, we run a Zermelo's probability model, PageRank and In-degree counting separately over it to rank the nodes, and compare the results.iii In our experimental comparisons, we apply these methods to data obtained from the Yelp social network.We find that PageRank and Indegree perform better than Zermelo's model in aggregation with paired preferences.Moreover, we also see that the paired preference model performs superior to simple rating sort ranking.Also, through an analysis of the proposed model we find a few important insights for Yelp platform like -how similar/dissimilar are the preferences in a group of friends, and what percentage of top ranked businesses have already been visited by a user.
Author: Rohit Bhoompally
Institution: University of Cincinnati
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2529675602