Abstract
The existing programs inside the voice assistant machine prompt human-machine interaction in response to a request from a user. However, the crucial problem is that the machine often may not give a proper answer to the user or cannot work out the existing program execution efficiently. Therefore, this study proposes a novel transform method to replace the existing programs (called sample programs in this paper) inside the machine with newly generated programs through code transform model GPT-2 that can reasonably solve the problem mentioned above. In essence, this paper introduces a theoretical estimation in statistics to infer at least a number of generated programs as required so as to guarantee that the best one can be found within them. In addition, the proposed approach not only imitates a voice assistant system with filtering redundant keywords or adding new keywords to complete keyword retrieval in semantic database but also checks code similarity and verifies the conformity of the executive outputs between sample programs and newly generated programs. According to code checking and program output verification, the processes can expedite transform operations efficiently by removing the redundant generated programs and finding the best-performing generated program. As a result, the newly generated programs outperform the sample programs because the proposed approach reduces the number of code lines by 32.71% and lowers the program execution time by 24.34%, which is of great significance.
Highlights
Alpha Go was developed by Google DeepMind in London in 2014, and it defeated all other Go masters
Four experiments are carried out in the following. e first experiment is to make word segmentation to select keywords and to optimize keyword retrieval. e second experiment is to search for sample programs and generate a number of preliminary programs based on the predetermined number of generated programs statistically. e third experiment is to analyze the pass ratio of code similarity checking and classify few preliminary programs as qualified programs
The keyword retrieval optimization was implemented in two aspects. e first one is to filter the redundant keywords, and the second is to add the required keywords. Evaluation metrics, such as accuracy, recall, and F1-Score, are used to measure the performance of keyword retrieval. e sample programs associated with keywords are obtained from GitHub [37] and both of them are stored in a semantic database
Summary
Alpha Go was developed by Google DeepMind in London in 2014, and it defeated all other Go masters. Research about human-computer interaction is meant to imitate human behavior, especially natural language representation and interpretation in the voice assistant machine [1]. Regarding the technology involved in the improvement method, a complete and quickly searchable semantic database using MariaDB is constructed with the Natural Language Toolkit (NLTK) model of sentence segmentation [5] to provide correct answers to users. The most fluent open source natural language model is the Generative Pre-trained Transformer 2 (GPT-2) [6], which is a natural language simulation machine developed by using OpenAI. This paper introduces a theoretical estimation in statistics to infer at least a number of generated programs produced by GPT-2, which guarantee that the best one can be found within them.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
