Towards Explainable Classification of Non-Functional Requirements Using Fine-Tune-Chain-of-Thought Reasoning

Supplementary material for Towards Explainable Classification of Non-Functional Requirements Using Fine-Tune-Chain-of-Thought Reasoning Fine-tuned model fine-tuned-flan-t5-base-nfr-identification-> Fine-tuned flan-t5-base model over 5 epochs for NFR identification fine-tuned-flan-t5-large-nfr-identification> Fine-tuned flan-t5-large model over 3 epochs for NFR identification fine-tuned-flan-t5-base-nfr-multilabel-classification -> Fine-tuned flan-t5-base model over 6 epochs for NFR classification fine-tuned-flan-t5-large-nfr-multilabel-classification-> Fine-tuned flan-t5-large model over 3 epochs for NFR classification Structure of artifacts.zip 1. Code 1. fine_tuning.py - This script is used for fine-tuning the classifiers. The same script works for both nfr identification and nfr classification (for training separate models). 2. inference.py - This script makes predictions using the finetuned models. It predicts on the datasets PURE-nfr-identification-data (level1).xlsx, PURE-nfr-identification-data (level1).xlsx, and PURE-nfr-classification-data (level2).xlsx in testind_dataset folder. 2. training_data 1. promise-exp-nfr-classification-data.xlsx - For training NFR classification model; Annotated in a multilabel classification scheme. 2. promise-exp-nfr-identification-data.xlsx - For training NFR identification model; Annotated in a multilabel classification scheme (FR, NFR, FR+NFR, None). Both of these files also contain the few-shot generated output of gpt-3.5 3. testing_data 1. EHR_nfr_identification (level 1).xlsx - Electronics Health Records dataset; Inference ran using fine-tuned flan-t5-large for NFR identification. 2. PURE-nfr-identification-data (level1).xlsx - PURE dataset with requirements annotated in a multilabel classification scheme (FR, NFR, FR+NFR, None); Inference ran using fine-tuned flan-t5-large and flan-t5-base for NFR identification. 3. PURE-nfr-classification-data (level2).xlsx - PURE dataset with NFRs annotated in a multilabel classification scheme.; Inference ran using fine-tuned flan-t5-large and flan-t5-base for NFR classification. Version 2 of this repository contains the llama generated output for NFR identification on level 1. Here is the link to version 2 > https://zenodo.org/records/10598270