People searching for people: analysis of a people search engine log

Recent years show an increas­ing inter­est in ver­ti­cal search: search­ing within a par­tic­u­lar type of infor­ma­tion. Under­stand­ing what peo­ple search for in these “ver­ti­cals” gives direc­tion to research and pro­vides point­ers for the search engines them­selves. In this paper we ana­lyze the search logs of one par­tic­u­lar ver­ti­cal: peo­ple search engines. Based on an exten­sive analy­sis of the logs of a search engine geared towards find­ing peo­ple, we pro­pose a clas­si­fi­ca­tion scheme for peo­ple search at three lev­els: (a) queries, (b) ses­sions, and © users. For queries, we iden­tify three types, (i) event-based high-profile queries (peo­ple that become “pop­u­lar” because of an event hap­pen­ing), (ii) reg­u­lar high-profile queries (celebri­ties), and (iii) low-profile queries (other, less-known peo­ple). We present exper­i­ments on auto­matic clas­si­fi­ca­tion of queries. On the ses­sion level, we observe five types: (i) fam­ily ses­sions (users look­ing for rel­a­tives), (ii) event ses­sions (query­ing the main play­ers of an event), (iii) spot­ting ses­sions (try­ing to “spot” dif­fer­ent celebri­ties online), (iv) poly­mer­ous ses­sions (ses­sions with­out a clear rela­tion between queries), and (v) repet­i­tive ses­sions (query refine­ment and copy­ing). Finally, for users we iden­tify four types: (i) mon­i­tors, (ii) spot­ters, (iii) fol­low­ers, and (iv) polymers.

Our find­ings not only offer insight into search behav­ior in peo­ple search engines, but they are also use­ful to iden­tify future research direc­tions and to pro­vide point­ers for search engine improvements.