Is Tofu the Cheese of Asia?: Searching for Corresponding Objects across Geographical Areas.

Keyword-based search engines are widely used nowadays for content retrieval. Creating queries is relatively easy when users wish to retrieve content in familiar domains (e.g., information about things within their own country). However, they often struggle when searching in unfamiliar domains (e.g., searching for information related to a foreign country). In this paper, we approach the vocabulary gap problem by allowing users to search by analogical examples, that is, by letting them utilize information in familiar domains to perform search in domains unfamiliar to them. In particular, we focus on geographical domains. We propose to build connections between two different spaces (e.g., USA and Japan) by mapping the distributed word representations in one space with the ones in the other space. We first introduce an effective technique for automatically constructing seed pairs of terms to be used for finding the optimal mapping function. Then we propose general and topic-based transformations of terms from one space to another. We test the performance of the proposed approaches on datasets derived from Wikipedia which are related to two quite diverse countries: Japan and USA.