Metadata-enriched Discovery Of Aspect Similarities Between Cultural Objects

Evaluating in the digital realm the similarity between two cultural objects relies on a number of factors. The latter can be subdivided to three categories, each describing a different set of properties. Specifically, the first such category pertains to generic attributes which usually are domain oblivious. For instance, for literature works standard text similarity metrics based on attributes such as n-gram or word frequencies. The second category consists of attributes deriving from domain-specific aspects. In the same example, a literature work may be described by the style, the author, and the year it was written. Finally, the third category relies on metadata generated by humans who have previously accessed the specific cultural object, perhaps drawn from a pre-specified list to avoid cluttering by redundant, erroneous, or invalid descriptions. Despite this restriction, the latter allows for an enhanced and more accurate description of the cultural object as a human can associate quicker and more efficiently a certain set of descriptions to it. Here a metric is proposed based on all three categories and it is compared against baseline ones with encouraging results.