Logo of the Information System for Agriculture and Food Research

Information System for Agriculture and Food Research

Information platform of the Federal and State Governments

SMAFIRA – Smart Feature-based interactive ranking (SMAFIRA)

Project

Food and consumer protection

This project contributes to the research aim 'Food and consumer protection'. Which funding institutions are active for this aim? What are the sub-aims? Take a look:
Food and consumer protection


Project code: BfR-TOX-08-1329-531
Contract period: 01.10.2014 - 30.01.2016
Purpose of research: Demonstration

Finding adequate animal replacement methods in literature databases is a difficult task, requiring great expertise, dedication and time. Literature databases are not optimized towards the search strategies needed, which demand close resemblance in goal and results, but a characteristic distinction in method (e.g. in vivo vs. in vitro). So, as a result, the result lists become cluttered with irrelevant publications, which have to be filtered out manually. This makes it nearly impossible in practice to decide whether a replacement method exists, and, if so, which one is the best. Yet, lawmakers demand that all experiments are conducted with the method most beneficial for the animals. In the SMAFIRA project, we work together with specialists from GESIS in machine learning. The goal is to provide a better automatic support for the search for animal replacement methods by learning from already established use cases.Goal of SMAFIRA is to improve the search for replacement methods and support the work of the BfR (consulting other agencies on this matter). In particular, we are interested in improving transparency and reproducibility of such results, thereby offering a standardized way of searching usable without any prior knowledge of animal replacement methods in particular. In this first stage of the project, the focus was to provide a proof-of-concept that automated methods can be used to significantly improve search. This included setting up the infrastructure that would allow us to objectively evaluate search success and to implement a variety of methods to find what would work best. There were two key conclusions. First, it possible, if difficult, to find suitable documents automatically, given a set of documents to train from. Second, it is possible to transfer the learning to other domains, thereby significantly improving the result lists that users get. We also tested un-supervised methods, but could not find any significant improvements with those we tested. However, we do believe there is some potential in them, e.g. by using hybrid approaches.

 

show more show less

Subjects

Framework programme

BMEL Frameworkprogramme 2008

Advanced Search