Recommending Recommendations to Support the Defense Acquisition Workforce

Abstract

This paper presentings the preliminary results of a research study to support the Defense Acquisition Workforce with a Natural Language Processing (NLP)/Machine Learning (ML) prototype of a system to determine what are the most relevant recommendations that stakeholders are providing to the Defense Acquisition community. The problem addressed by the research study is in the realm of NLP and ML and it is part of the quite popular category of “recommendation systems.” Unlike the majority of the cases in this category, though, this task does not focus on numerical data representing behaviors (like in shopping recommendations), but on extracting user-specific relevance from text and “recommending” a document or part of it. In order to identify important pieces of these texts, subjective text analysis is required to be run. The method used for the analysis is the “room theory framework” by Lipizzi et al. (2021) which applies the Framework Theory by Marvin Minsky (1974) through the use of text vectorization. This framework has three main components: a vectorized corpus representing the knowledge base of the specific domain (the “room”), a set of keywords or phrases defining the specific points of interest for the recommendation (the “benchmarks”) and the documents to be analyzed. The documents are then vectorized using the “room” and compared to the “benchmarks.” The sentences/paragraphs within a given document that are most similar to the benchmarks, and thus presumably the most important parts of the document, are highlighted. This enables the DAU reviewers to submit a document, run the program, and be able to clearly see what recommendations will be the most useful.