AI Chatbot Engagement Dataset

Only reviews containing education‑related keywords (e.g., homework, study, exam, student, teacher) were retained, resulting in a final sample of 83,752 English‑language reviews after text cleaning, removal of reviews shorter than 5 words, and language detection. Variables Column Description score User rating (1 to 5 stars) App Platform identifier: ChatGPT or DeepSeek cognitive Binary indicator (1 = review mentions explanation quality, step‑by‑step reasoning, accuracy, etc.) affective Binary indicator (1 = review mentions encouragement, patience, supportive tone, etc.) social Binary indicator (1 = review mentions conversational naturalness, human‑likeness, warmth, etc.) ethical Binary indicator (1 = review mentions trust, privacy, bias, fairness, etc.) processed_text Lemmatised, stopword‑removed, lowercased text of the review (after cleaning) How the Binary Indicators (Cognitive, Affective, Social, Ethical) Were Derived Each binary indicator was created using dictionary‑based keyword matching on the processed_text field, with whole‑word matching (regular expression \bword\b).

A review was coded as 1 for a given factor if any keyword from the corresponding dictionary appeared in its processed_text. The keyword dictionaries are: Cognitive keywords explain, explanation, step by step, step, detail, depth, accurate, accuracy, correct, reasoning, logic, understand, understanding, clarify, clear, comprehensive, thorough, insight, knowledge, learn, learning, teach, teaching, tutor, homework, assignment, problem solving, solution, answer, question, math, science, coding, programming, essay, writing, grammar Affective keywords encouraging, encouragement, motivate, motivating, motivation, support, supportive, patient, patience, kind, friendly, helpful, understanding, empathetic, empathy, care, caring, nice, pleasant, positive, encouraged, inspired, confidence, frustrat, annoying, angry, upset, disappointed Social keywords conversation, conversational, natural, human, humanlike, like a friend, companion, interact, interaction, dialogue, chat, talk, speak, respond, response time, engaging, personable, warm, cold, robotic, mechanical Ethical keywords trust, trustworthy, reliable, bias, biased, fair, unfair, privacy, private, data, secure, safety, transparent, transparency, honest, manipulate, manipulative, ethical, ethics, responsible, accountability, mislead, misinformation, fake Reviews were filtered for education keywords, cleaned (URLs, non‑letters, lowercase), kept ≥5 words, detected as English (langdetect), then tokenised, stopword‑removed, lemmatised (NLTK), and trimmed to ≥3 characters.