AI Text Classifier

Create an AI Text Classifier for customer feedback analysis. Classify each review, survey response, or social media comment into predefined sentiment and topic labels. Use the following requirements: Objective: Automatically label customer feedback so product, support, and marketing teams can identify trends and prioritize action. Input text examples may include product reviews, NPS comments, support feedback, app store reviews, and social media mentions. Classification tasks: 1. Sentiment classification: positive, neutral, negative, mixed. 2. Topic classification: pricing, product quality, usability, delivery, customer support, billing, feature request, bug report, cancellation risk, general praise, other. 3. Urgency classification: low, medium, high. 4. Multi-label support: allow more than one topic when relevant. Output format for each text: - original_text - sentiment_label - sentiment_confidence_score - topic_labels - topic_confidence_scores - urgency_label - urgency_confidence_score - short_reasoning - recommended_business_action Instructions: - Prioritize precision for negative feedback and cancellation-risk labels. - If the text is ambiguous, assign the most likely label and mark confidence as low. - Detect language if possible and include the language code. - Do not invent information not present in the text. - Return results in valid JSON. Also provide: 1. A recommended label taxonomy. 2. 10 sample labeled examples. 3. Evaluation metrics to track, including precision, recall, F1, and confusion matrix. 4. Suggestions for improving accuracy with custom training data.

A production-ready customer feedback classification workflow that labels sentiment, topics, urgency, and recommended business actions in structured JSON. The result helps teams identify negative trends, detect churn risk, summarize customer pain points, and prioritize product or support improvements.

Create an AI Text Classifier for routing inbound customer support tickets. The classifier should identify user intent, urgency, department, and whether human escalation is required. Objective: Automatically classify incoming support emails, chatbot messages, and contact form submissions so tickets are routed to the correct team faster. Input text may include short or long customer messages with incomplete information. Classification tasks: 1. Intent label: password_reset, refund_request, billing_question, technical_issue, account_access, order_status, cancellation_request, complaint, feature_request, sales_inquiry, security_concern, other. 2. Department routing label: support, billing, engineering, sales, security, customer_success, operations. 3. Priority label: low, normal, high, urgent. 4. Escalation label: no_escalation, human_agent_required, manager_required, security_team_required. 5. Sentiment label: positive, neutral, frustrated, angry, confused. Output format for each ticket: - ticket_text - detected_language - intent_label - intent_confidence - routing_department - routing_confidence - priority_label - priority_confidence - escalation_label - sentiment_label - explanation - suggested_agent_reply_summary Rules: - Classify security threats, payment failures, account lockouts, and angry cancellation messages as high or urgent depending on severity. - If the classifier is uncertain, route to support and set confidence below 0.60. - Use multi-label intent only if the message clearly contains multiple requests. - Return valid JSON only. Also include: 1. A decision logic summary. 2. Example training dataset schema. 3. 15 synthetic labeled examples. 4. Recommended production monitoring plan for drift, misroutes, and low-confidence predictions.

A support ticket routing classifier specification that assigns intent, department, priority, escalation need, and sentiment for each inbound message. The result enables faster ticket triage, fewer misroutes, consistent escalation decisions, and better customer response workflows.

Create an AI Text Classifier for content moderation. The classifier should detect whether user-generated text violates platform policy and assign clear risk categories. Objective: Automatically classify posts, comments, usernames, messages, and reviews to support moderation decisions while reducing false positives. Classification tasks: 1. Moderation category: safe, spam, harassment, hate_speech, sexual_content, self_harm, violence_threat, illegal_activity, personal_data, scam_or_phishing, misinformation, profanity, other_policy_violation. 2. Severity level: none, low, medium, high, critical. 3. Action recommendation: allow, allow_with_warning, hide_pending_review, remove, escalate_to_human, escalate_to_trust_and_safety. 4. Confidence score for each assigned category. 5. Explanation focused on policy signals, not moral judgment. Output format for each text: - content_text - detected_language - moderation_categories - category_confidence_scores - severity_level - recommended_action - requires_human_review - policy_reasoning - safer_rewrite_if_applicable Rules: - Prioritize recall for high-risk categories such as self-harm, credible violence threats, personal data exposure, and phishing. - Prioritize precision for hate speech and harassment to reduce unfair removals. - If context is insufficient, mark requires_human_review as true. - Do not repeat sensitive personal data in the explanation; summarize it instead. - Return valid JSON only. Also provide: 1. A moderation label taxonomy with definitions. 2. 20 example inputs with expected labels. 3. Evaluation approach using precision, recall, F1, false positive rate, and false negative rate. 4. Human-in-the-loop review strategy for uncertain or high-severity cases.

A content moderation classifier workflow that categorizes user-generated text by policy risk, severity, confidence, and recommended action. The result supports scalable moderation, human review prioritization, safer community management, and auditable policy-based decisions.