Social Network Search as a Volatile Multi-armed Bandit Problem Zahy Bnaya Information Systems Engineering Department Ben Gurion University, Beer Sheva, Israel zahy@bgu.ac.il Rami Puzis Information Systems Engineering Department Ben Gurion University, Beer Sheva, Israel puzis@bgu.ac.il Roni Stern SEAS Harvard University, MA, USA roni.stern@gmail.com Ariel Felner Information Systems Engineering Department Ben Gurion University, Beer Sheva, Israel felner@bgu.ac.il ABSTRACT In many cases the best way to ﬁnd a proﬁle or a set of proﬁles matching some criteria in a social network is via targeted crawling. An important chal- lenge in targeted crawling is choosing the next proﬁle to explore. Existing heuristics for targeted crawl- ing are usually tailored for speciﬁc search criterion and could lead to short-sighted crawling decisions. In this paper we propose and evaluate a generic ap- proach for guiding targeted crawling which is based on recent developments in Artiﬁcial Intelligence. Our approach, based on the recently introduced variant of the Multi-Armed Bandit problem with volatile arms (VMAB), aims to provide a proper balance between exploration and exploitation during the crawling pro- cess. Unlike other heuristics which are hand tailored for speciﬁc type of search queries, our approach is general-purpose. In addition, it provides provable performance guarantees. Experimental results indi- cate that our approach compares favorably with the best existing heuristics on two diﬀerent domains. I INTRODUCTION Online social networks (SNs) such as Facebook, Twit- ter, Google+, or Academia.edu are a part of every- day life for many people around the world. SNs are a source of valuable information. This information can be used for tracking public opinions, political trends, disseminate news, or just socialize by ﬁnding the right group of people to communicate with. Extracting information from SNs, referred to as SN querying, is widely practiced by commercial compa- nies, government agencies and even individual people. The objective of SN queries is to pinpoint proﬁles that provide valuable information according to some cri- teria. SN queries can be used to select subjects for surveys, research, marketing purposes etc. For example, consider an SN query designed to help the chairs of a conference program to assemble pro- gram committee and/or ﬁnd reviewers. Meeting the criteria for this query may be to match proﬁles of re- searchers that had written a number of papers in a particular ﬁeld and participated in previous confer- ences. Another possible example is selecting from a group of known candidates, the ones that are most inﬂuential or lead public opinions in a certain mat- ter of interest, such as a speciﬁc consumer’s product or political issues. These individuals can be found by examining followers on Twitter or Facebook and checking the topic of the tweets or posts on which the followers are active. Proﬁles that are active on the speciﬁc matter and tend to be followed can be regarded as satisfying this query criteria. Lately, Facebook realized the need for SN queries and introduced the SN search feature. However, many other SN services lag behind and do not provide search capabilities at all. Although very intuitive, the search capabilities currently provided by Face- book are very limited. In many cases a complex cri- teria is required to deﬁne which proﬁles are relevant and which are not. The search capabilities provided by Facebook are not suﬃcient to execute complex searches where the relevance of a proﬁle can only be evaluated by deep analysis on the proﬁle content. In case of complex SN queries, the relevant proﬁles are commonly pinpointed using dedicated crawls on the SN. We assume that during such crawls a function for analyzing a proﬁle and evaluating its relevance, referred to as proﬁle-acquisition, is repeatedly exe- cuted in order to: 1) Extract information from the analyzed proﬁles and 2) Extract new proﬁles from their lists-of-friend (LOF) which become candidates for future proﬁle-acquisition. The aim of this paper is to provide eﬃcient policies for guiding the SN crawl- ing process by intelligently choosing the next proﬁle to acquire. The challenge is to focus the search at relevant areas of the network that are most likely to contain valuable proﬁles which will satisfy the criteria of the query. Previous work has shown that blind, arbitrary selec- tion of next proﬁles to examine (aka. ”Snowball“ crawling) is ineﬃcient [1]. Thus, a number of intelli- gent selection methods for the next proﬁle to acquire Page 1 of 15 c ASE 2012