Identifying problems and solutions in scientific text

Abstract

Research is often described as a problem-solving activity, and as a result, descriptions of problems and solutions are an essential part of the scientific discourse used to describe research activity. We present an automatic classifier that, given a phrase that may or may not be a description of a scientific problem or a solution, makes a binary decision about problemhood and solutionhood of that phrase. We recast the problem as a supervised machine learning problem, define a set of 15 features correlated with the target categories and use several machine learning algorithms on this task. We also create our own corpus of 2000 positive and negative examples of problems and solutions. We find that we can distinguish problems from non-problems with an accuracy of 82.3%, and solutions from non-solutions with an accuracy of 79.7%. Our three most helpful features for the task are syntactic information (POS tags), document and word embeddings.

Scholarly data mining: A systematic review of its applications

“…Document structure analysis (Boyack et al, 2018; Heffernan & Teufel, 2018; Lu et al, 2018) studies the internal document structure by identifying the functional structure (further detailed at three levels: section header‐based identification, section content‐based identification, or paragraph‐based identification [Lu et al, 2018]), identification of problems and solutions in a specific paper by making a binary decision about problem‐hood and solution‐hood of a given phrase in article Heffernan and Teufel (2018), or by studying research proposals and analyzing their discourse for clarity (Boyack et al, 2018). For the functional structure identification, the authors in Lu et al (2018) have proposed a novel clustering algorithm to generate a domain‐specific functional structure, applied to 300 research articles in computer science.…”

“…The application of the proposed approach, in two tasks: academic search and keyword extraction, confirms that the identified structure obtains more relevant information and achieves better performance. However, for the identification of problems and solutions, the authors in (Heffernan & Teufel, 2018) have proposed an automatic classifier that makes a binary decision about problem‐hood and solution‐hood of a given phrase, that may or may not be a description of a scientific problem or a solution. The authors have defined a set of 15 features, including syntactic information (POS tags), document and word embeddings, and have applied several ML algorithms such as Naïve Bayes, Logistic Regression and Support Vector Machine, on a corpus of 2000 positive and negative examples of problems and solutions extracted from the 2016 ACL (Association of Computational Linguistics) anthology.…”

Word Embedding for Rhetorical Sentence Categorization on Scientific Articles

“…Heffernan & Teufel [17] employed Word2Vec representation to identify problem statements in scientific texts. They used 18,753,472 sentences from a biomedical corpus consisting of all full-text Pubmed articles and then built a model from 200 words that are semantically similar to 'problem'.…”

Co-evolution of COVID-19 Research and China’s Policies

“…As scientific research is a process of problem-solving and resolving disputes (Heffernan and Teufel 2018;Fortunato et al 2018), publications on COVID-19 indirectly reflect whether a scientific consensus has been reached, or whether a solution has been developed for a particular problem. Such connections can be revealed when one studies the length of time that particular topics remain popular.…”