Lattice LSTM Model for Function Point Based Software Cost Measurement
Ming Qin · 2019
One of the core problems in the field of software engineering is to measure the software development cost, which helps to ensure the software quality and avoid the waste of investment. Traditionally, the software cost measurement is implemented by experts who are strictly trained, which is relatively expensive and inefficient. In this paper, a neural network based machine learning method is proposed to perform software cost measurement task, which decreases the time and manpower costs. The proposed method takes advantage of the long short term memory (LSTM) neural network and conditional random field to identify different types of function points in software requirements document, where the function points is the key to measure to software cost. To solve the incorrect word segmentation problem in open domain such as software development cost field, lattice LSTM model is constructed to utilize both Chinese characters and words information. Experiments show the proposed method significantly reduces the human labor and ensure the quality of software cost measurement.