On the use of Bayesian networks to combine raw data from related studies on sensory satiation

详细信息	查看全文 | 推荐本文 |

• 作者：V.A. Phan^a ; ^{van-anh.phan@wur.nl} ; ^{contact.vananh.phan@gmail.com} ; M.G. Ramaekers^a ; ^{marielle.ramaekers@wur.nl} ; D.P. Bolhuis^a ; ^{dieuwerke.bolhuis@wur.nl} ; U. Garczarek^b ; ^{Ursula.Garczarek@unilever.com} ; M.A.J.S. van Boekel^a ; ^{Tiny.vanBoekel@wur.nl} ; M. Dekker^a ; ^{Matthijs.Dekker@wur.nl}
• 关键词：Bayesian networks ; Combine raw data ; Structural linking information ; Global experimental design ; Ad libitum intake ; Sensory satiation
• 刊名：Food Quality and Preference
• 出版年：2012
• 期刊代码：52_09503293
• 类别：abs
• 出版时间：October, 2012
• 卷：26
• 期：1
• 页码：119-127
• 文件大小：902 K

摘要

Bayesian networks were used to combine raw datasets from two independently performed but related studies. Both studies investigated sensory satiation by measuring ad libitum intake of a tomato soup model. The Aroma study varied aroma concentration and aroma duration as the explanatory variables, and the Taste study varied salt intensity. To combine the data from the two studies, the Aroma study needed information on salt aspects for all of its observations. Equally, the Taste study needed information on aroma aspects. This information was used to link the two single networks, each representing one study, into a combined network; therefore, it is referred to as structural linking information. The approach taken is seen as an example for the potential benefit and the challenges when combining raw datasets from different studies. The combined network is able to generate additional insights into complex relationships encountered with research on satiation. The main challenge results from the missing of structural linking information. In this paper, we (1) suggest solutions for obtaining the structural linking information, and (2) propose an approach to global experimental design to prevent this situation. The nature of the paper is theoretical rather than analytical due to the limitations caused by the small size of datasets.