This paper describes a demonstration of the Discovery Hub exploratory search system. The demonstration focuses on the exploration of topics through several perspectives. Exploratory search refers to cognitive-consuming search tasks like learning or investigation. There is a need to develop systems optimized for supporting exploratory search as the today widely-used search engines fail to e ciently support exploratory search [1]. Linked data o ers exciting perspectives in this context and several systems were already published. We want to reach a new level in linked data based exploration by allowing the users to unveil knowledge nuances corresponding to speci c facets of interest of the topic explored. In this demonstration paper we rst give a brief overview of linked data based exploratory search systems. Then we present the Discovery Hub web application and focus more particularly on its multi-perspectives exploration capacity. Finally we present the demonstration scenario we propose.
The amount of contributions at the crossroad of semantic search and exploratory search is increasing today. The linked data knowledge, and especially DBpedia, allows to design new information retrieval approaches and interaction models that e ciently support exploratory search tasks. Yovisto3 (2009) is an academic videos platform that retrieves topics' suggestions that are semantically related to the users' query. The objective is to ease the exploratio of the videos collection. Lookup Explore Discover4 (2010) helps the users to compose queries about topics of interest by suggesting related query-terms. Once the query is composed the system retrieves the results from others several services such as search engines and social networks. Aemoo5 (2012) o ers a graph-view on topics of interest. The graph shows their neighborhood ltered by a semantic pattern.
The users can reverse the ltering to show more surprising knowledge. They can also ask for explanations (cross-references in Wikipedia) about the relations between the shown resources. The Seevl6 (2013) demonstrator is a music discovery platform implementing a linked data based recommendation algorithm. The DBpedia semantics are also used in Seevl to support browsing (e.g. by music genres, band members) and to provide explanations about the recommendations (showing the shared properties between the artists). Linked Jazz7 (2013) aims to capture the relations within the American jazz community in RDF. The authors rely on a large amount of jazz people interviews transcripts. These transcripts are automatically processed and then nely analyzed through a crowd-sourced approach.
The approaches recently published in the literature produced good results when evaluated. Nevertheless a common limit of the existing linked data based exploratory search systems is the fact they constrain the exploration through single results selection and ranking schemes. The users cannot in uence the retrieved results to reveal speci c aspects of knowledge that interest them in particular. 3
The framework and models implemented by the Discovery Hub application was
presented in [
The fact the results are computed at query-time allows to let the users control several computation parameters through the interface and to o er multiperspective exploratory search. Indeed, the objects described in linked data datasets can be rich, complex and approached in many manners. For example, a user can be interested in a painter (e.g. Claude Monet or Mary Cassat ) in many ways: works, epoch, movement, entourage, social or political contexts and more. The user may also be interested by basic information or unexpected ones depending on his actual knowledge about the topic. He may also want to explore the topic through a speci c culture or area e.g. impressionism in American or French culture.
7 http://linkedjazz.org/ 8 http://discoveryhub.co, current CPU-intensive experiments might slow down the search temporarily 9 https://www.youtube.com/user/wearediscoveryhub/videos
The framework allows three operations for building such exploration
perspectives, detailed in [
During the results examination we will voluntarily focus on the French impressionist painters that were close to Monet. At this point the user might be interested in the relations of Monet with the non-French impressionists (query 2 in Table 1). We will explain the querying system for criteria of interest speci cation and then emphasize the di erences between the results obtained with query 1 and 2.
To continue in the same logic we will submit the query 3 as well as a query with a high level of randomness and one using the French chapter of DBpedia in several tabs. We let the audience compare the results. We seek an interactive demonstration by encouraging the audience to try the application while commenting the system more than a strict and pre-de ned sequence of interactions (which serves only to start the interactions). 5
Discovery Hub is a linked data based exploratory search system built on the top of DBpedia. With this demonstration we want to show the value of linked data for exploratory search. Mature datasets like DBpedia allow the creation of new information retrieval approaches as well as new interaction models. More speci cally we want to demonstrate the multi-perspectives exploratory search capacities of Discovery Hub. Thanks to the demonstration track we hope to have discussions with other researchers about the perspectives we envision for Discovery Hub. It notably includes an approach where the user can specify or change the speci ed criteria of interest interactively in order to re-rank the results without relaunching the whole query-process.