Recognizing Named Entities in Failure Analysis Reports

Failure Analysis (FA) is a complex laboratory activity that requires systematic documentation of all findings and conclusions obtained during an analysis to preserve knowledge acquired by engineers in this process. Current FA information systems store this data in different formats distributed across databases, file shares, wikis, or other human-readable forms. Given a large volume of generated FA data, navigating or searching for particular information is hard since machines cannot automatically process the stored knowledge and require frequent interaction with experts. This paper investigates two applications of modern Natural Language Processing (NLP) approaches to the Named Entity Recognition (NER) task. In particular, we study the performance of two techniques, spaCy, a highly regarded Python library, and the state-of-the-art BERT Language Models (LM), pretrained on semiconductors data. Our experiments show that spaCy reached precision, recall, and F1 scores of about 23%, whereas a BERT-based model achieved a precision of 51%, recall of 49%, and an F1 score of 50% on the test corpus.