Article Preview
TopIntroduction
Knowledge and semantics are needed both in quality and quantity. Many contemporary applications rely on (semantic) metadata in order to provide their intended functionality (Siorpaes & Simperl, 2010). Consequently, the creation or acquisition of such metadata is crucial to their effective operation and ultimately user satisfaction. Knowledge representations, from formal ontologies to lightweight taxonomies and flat folksonomy-like term networks, are especially vital to advanced information processing tasks that need to process semantic relationships between entities. Typical examples of applications are metadata-based search engines or faceted browsers (Tvarožek & Bieliková, 2010), which require either document annotations or faceted classifications, (personalized) e-learning systems (Barla et al., 2010), which require complex course metadata, information repositories that need resource interlinks and annotations (e.g., Wikipedia), or Semantic Web applications that require formal ontologies. The use of concept relationships is also widely used in exploratory search tasks (Marchionini, 2006) like search query expansion (Ungrangsi, Anutariya, & Wuwongse, 2010) or visualization and navigation in information spaces (Stewart, Scott, & Zelevinsky, 2008).
Rather than use of semantics, more problematic is their acquisition. While automated approaches are able to provide quantity, in comparison with human oriented approaches (expert work, crowdsourcing) they vary in the quality of semantics they provide. The concept of games with a purpose (GWAP) emerged within the human computing initiative in recent years and stresses the use of human problem-solving capabilities via specially engineered games to address so called human intelligence problems (HITs) that are currently too difficult to solve by machine approaches (e.g., image annotation) (Siorpaes & Hepp, 2008a). By transforming computational problems into engaging computer games, GWAPs enable us to take advantage of human computing power without having to pay for expensive human resources while also providing the scalability to web-scale tasks. Several successful games like the image annotation ESP Game (von Ahn & Dabbish, 2008) have already shown the potential of this approach, especially for creation of web semantics (Siorpaes & Hepp, 2008a).
However, the creation of GWAPs is not a straightforward process, because it is specific to each human intelligence problem that it tries to solve. Although some effort has been spent on developing generic methodologies for GWAP creation (von Ahn & Dabbish, 2008; Siorpaes & Hepp, 2008a; Vickrey et al., 2008), they remain applicable only to narrow problem domains. In this paper, we examine design aspects of existing GWAPs (e.g., player input validation, anti-cheating, scoring system) – their shared characteristics and specifics to support the effort of devising a broader methodology, at least for the Semantic Web domain.
Our own contribution to the field of GWAPs for the Semantic Web includes Little Search Game – a novel game for the discovery of semantic links between terms and the successive construction of a term network. Little Search Game is a web search game, where players compete in reducing the number of search results by entering queries in a special format. The query consists of one normal search term, given to players before the game starts, and several negative terms entered by players, like “star –movie –wars –death”. This query format forces players (in order to be successful in the game) to use negative terms with frequent common co-occurrences with the initial term (i.e., related to it). We collect this information via game logs and aggregate it into a lightweight term network of term relationships. The game has also the unique ability to discover term relationships as perceived by humans, some of which are hard to detect using statistical corpora analysis. Therefore the resulting lightweight term network is suitable for applications like learning support frameworks (Barla et al., 2010), exploratory search tools (Šimko, Tvarožek, & Bieliková, 2010) or as a subject for further semantic enrichment.