Causal Relation Extraction from Failure Analysis Documents

Nobuyuki Ohmori and Tatsunori Mori


Product development, tangibility


In product development, because product reliability requirements are often not fully satisfied, the prevention of product failure has become a matter of major concern. In order to support the development of reliable products, we propose information extraction methods that extract causal information from product failure documents. These methods use a binary classifier that is trained on the basis of lexico-syntactic patterns and a thesaurus, and we implement the binary classifier by using a machine learning system. In order to improve the accuracy of extraction, we introduce the notion of “tangible words.” The experimental results show that the proposed methods offer improved precision over a previous method. These methods help to prevent product failure by using information about failure and the causes of the failure that are related to products under development.

Important Links:

Go Back