Design rationale clustering method based on density peaks

To solve the design rationale automatic clustering problem, a design rationale clustering method by using density peaks clustering was proposed. Combined with the semantic features of design rationale, the design rationale was transformed into the feature vector with TF-IDF method. The local density and distance of each vector was obtained based on density peaks clustering. These two parameters were expressed as the decision graph, and the remaining data points were assigned to the same cluster as its nearest neighbor of higher density. Aiming at the problem that density peaks clustering could not deal with data of uneven distribution, the dynamic cut-off distance was defined to improve the local density function. 55 design rationale examples from a mechanical design team were illustrated to prove the effectiveness of the proposed method.