Acceptance Evaluation of Code Recommendation Systems by Programming Behaviors Detection and Analysis

摘要

Code recommendation system is used for searching and recommending code for programmers. Traditional evaluation code recommendation system approaches are commonly based on questionnaire surveys. However, empirical studies have indicated that end-users are usually unwilling to fill out surveys, which may interrupt their programming work. In this paper, we have proposed a new evaluation developers' acceptance approach whenever a source code document is recommended, called Acceptance Automatic Analyze Approach Based on Programming Behaviors(4APB). The evaluation mechanism works automatically by detecting and analyzing programming behaviors via an IDE plug-in, without additional actions from developers. The acceptance analyzation solution is based on innovation Extractive and Exact Model (EEM). It's a special two layers GRU[1]core model for programming behaviors. The first GRU layer extract general thought of programmers, and the second one calculates the exact acceptance. According to the result of experience, EEM shows a high accuracy.