Design and Optimization of an Automatic Mobile Application Generating Learning Platform

EasyChair Preprint 9136

Abstract

We present a large language model-based learning platform that let students automatically generate mobile applications for smartphones and tablets from natural language descriptions. Furthermore, we show that the user-generated apps can be optimized with simple modifications to the generative model’s input (”prompts”). This paper explores three different methods of modifying the prompt:1) changing the selection mechanism of example pairs,2) varying the number of example pairs, and 3) altering how the pairs are ordered within the prompt. Prompts are constructed from a set of example pairs(a textual description of an example app and its corresponding code) along with the description of the desired app. We evaluated the model’s performance with 18 possible candidate mobile apps, ranging from simple to complex, and used the BLEU score to compare the outputs to manually created apps. Our results show that appropriate example pair selection and variation of the number of example pairs make a difference in the quality of the generated apps, but alteration of example pair ordering does not. We conclude with a discussion about the potential implications for CS education in light of generative models for code.

Keyphrases: Intelligent Learning Platform, Prompt Engineering, large language models, mobile application

@booklet{EasyChair:9136,
  author    = {Jasmine L Shone and Robin Liu and Evan Patton and David Young-Jae Kim},
  title     = {Design and Optimization of an Automatic Mobile Application Generating Learning Platform},
  howpublished = {EasyChair Preprint 9136},
  year      = {EasyChair, 2022}}