Inverted file-based general metric space indexing for quality-aware similarity search in information retrieval

Abstract

The notion of quality in its broadest sense is central to information retrieval (IR) where a user's information need is to be fulfilled as good as possible. A user searching for cars on sale in Bamberg might be interested in car dealers geographically close to Bamberg with high user ratings. The buyer might already know or trust a person who trusts the particular dealer. Furthermore, the cars which are sold by the dealer should offer a high quality on different levels-the type of car in general as well as the car to be bought. If the buyer can only travel to Bamberg on weekends, availability of the car dealer becomes another important factor. As this example shows, the integration of various quality aspects in IR is challenging but essential. Thus, there is a need for scalable and efficient indexing and retrieval techniques which can cope with such search situations. Here, metric space access methods (MAMs) present a flexible indexing paradigm.We will briefly review these techniques and show how they can be applied in the context of qualityaware IR. Furthermore, we will present IF4MI which is purely based on the inverted file concept and thus inherently provides a multi-feature MAM. It can make use of extensive knowledge in the field of inverted file-based indexing and represents a versatile indexing technique for quality-aware IR. © Springer-Verlag Berlin Heidelberg 2013.