Evaluation of RFPs Based on Machine Learning
Yasuhiro Saito, Akito Monden, Ken‐ichi Matsumoto · 2013
This paper proposes a machine learning approach to evaluate the clarity of non-functional requirements (NFRs) described in a Request For Proposal (RFP) written in a natural language. In the proposed method, keywords related to NFRs are extracted from a RFP, and mapped to each NFR category. Then, the clarity of NFRs is modeled by the random forest with weight factors based on appearance frequency and context vectors. As a result of an experimental to evaluate the clarity (low, mid or high) of many NFR categories in 70 RFPs, the proposed method showed 69.8% match to the expert’s decision. Also, there were few cases where the model concluded as clarity=high while expert concluded clarity=low, and vice versa. These results suggest that the proposed machine learning approach could be used to automatically evaluate the quality of RFP without experts.