Supporting data for "Using Artificial Intelligence to Promote Productive Talk: Automatic Identification, Prediction Explanation, and Practical Evaluation"

Scope of the Research Data:This data collection supports a doctoral thesis investigating the integration of Artificial Intelligence (AI) for analysing classroom dialogue and supporting teacher professional development in dialogic pedagogy. The research comprises four interrelated studies: 1) a systematic review of the existing literature; 2) the development and evaluation of AI models for classifying productive "talk moves" in classroom discourse; 3) a quasi-experiment examining how explainable AI affects teacher trust and acceptance; and 4) a comparative study evaluating the educational impact of different AI models within a professional development programme. The primary target participants across the experimental studies were pre-service teachers (total n=119 across two studies). The data reflects a mixed-methods approach, containing bibliographic records, quantitative model performance metrics, Python source code, anonymised participant questionnaire responses, assessment results, and anonymised interview transcripts.Description of Data Files:The data is organised by the chapter of the thesis in which it is primarily featured.Chapter 3 Data: This chapter contains the data for the systematic review (Study 1). The folder exported_data_from_keywords_search holds the exported reference records from the Web of Science and EI Compendex databases, which formed the initial corpus for screening. The Excel file coded_data_for_included_studies.xlsx contains the final coded data extracted from the 68 studies included in the review, encompassing bibliographic information, educational contexts, technical methodologies, and reported outcomes.Chapter 4 Data: This chapter contains the data for developing AI models for talk move analysis (Study 2). The folder Cleaned_TalkMoves_dataset contains a publicly available dataset of classroom dialogue transcripts, which was cleaned and prepared for model training and evaluation. The codes folder contains all original Python scripts written for the project, including those for data preprocessing, implementing Parameter-Efficient Fine-Tuning (PEFT) of Large Language Models (LLMs), model training, and performance evaluation.Chapter 5 Data: This chapter contains data for the quasi-experiment on explainable AI and teacher trust (Study 3). The experimental_group folder holds the anonymised data from participants who interacted with an AI tool enhanced with explainability features. This includes their questionnaire responses measuring trust, technology acceptance, and cognitive load, as well as relevant test data.Chapter 6 Data: This chapter contains data for the professional development impact study (Study 4). The control_group folder holds the anonymised parallel data from participants in the comparative condition. Both the experimental and control group folders contain anonymised questionnaire data on pedagogical knowledge and learning motivation, pre/post-test data, and anonymised transcripts of post-study interviews conducted with participants.All participant data has been anonymised to protect confidentiality. The dataset as a whole provides the underlying evidence for the thesis's investigations into the technical, human-interaction, and pedagogical dimensions of using AI to support the analysis and practice of classroom dialogue.