|Title||Characterizing Concepts in Taxonomy for Entity Recommendations|
|Year of Publication||2016|
|Authors||Siva Kumar Cheekula|
|Academic Department||Department of Engineering & Computer Science|
|Number of Pages||59|
|University||Wright State University|
Entity recommendation systems are enormously popular on the Web. These systems harness manually crafted taxonomies for improving recommendations. For example, Yahoo created the Open Directory Project for search and recommendation, and Amazon utilizes its own product taxonomy. While these taxonomies are of high quality, it is a labor and time-intensive process to manually create and keep them up to date. Instead, in this era of Web 2.0 where users collaboratively create large amounts of information on the Web, it is possible to utilize user-generated content to automatically generate good quality taxonomies. However, harnessing such taxonomies for entity recommendations has not been well explored. We exploit the Wikipedia category structure as a taxonomy and explore three prominent characteristics of concepts in the taxonomies for entity recommendations. The three characteristics we explore are: 1) Specificity, 2) Priority, and 3) Domain Relatedness of concepts in the taxonomy. We demonstrate the utility of specificity and priority of concepts in the taxonomies in achieving high quality recommendations by evaluating our recommender system on two diverse datasets.