Deceptive Review Detection

# Deceptive Review Detection
# Based on Ott et al., ACL 2011
# Paper: https://aclanthology.org/P11-1032/
#
# This task distinguishes truthful reviews (genuine experiences)
# from deceptive reviews (fake, written to mislead).
#
# Key Findings from Research:
# - Humans perform at chance level (50%) at detecting fake reviews
# - Deceptive reviews contain more verbs, adverbs, pronouns
# - Truthful reviews contain more nouns, adjectives, details
# - Fake reviews often set the scene (vacation, business trip)
# - Truthful reviews focus on specific hotel features
#
# Linguistic Patterns:
# Truthful reviews tend to be:
# - More specific about features (room, bathroom, bed)
# - More concrete and detailed
# - More nouns and spatial information
#
# Deceptive reviews tend to:
# - Use more superlatives and exaggeration
# - Focus on why the reviewer was there
# - Be more narrative/story-like
# - Use more first-person pronouns
#
# Annotation Guidelines:
# 1. Read the full review carefully
# 2. Look for specificity vs vagueness
# 3. Consider: Does this feel like lived experience?
# 4. Watch for excessive praise or generic descriptions
# 5. Note: This is difficult - humans struggle at this task

annotation_task_name: "Deceptive Review Detection"
task_dir: "."

data_files:
  - sample-data.json
item_properties:
  id_key: "id"
  text_key: "review"

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

annotation_schemes:
  # Step 1: Authenticity classification
  - annotation_type: radio
    name: authenticity
    description: "Is this review truthful or deceptive?"
    labels:
      - "Truthful"
      - "Deceptive"
      - "Uncertain"
    tooltips:
      "Truthful": "The review appears to describe a genuine experience"
      "Deceptive": "The review appears to be fake or fabricated"
      "Uncertain": "Cannot determine with reasonable confidence"

  # Step 2: Deception indicators (if applicable)
  - annotation_type: multiselect
    name: indicators
    description: "Which indicators influenced your judgment? (Select all that apply)"
    labels:
      - "Too generic/vague"
      - "Excessive superlatives"
      - "Narrative/story-like"
      - "Lacks specific details"
      - "Focuses on reviewer not product"
      - "Specific concrete details"
      - "Mentions negatives honestly"
      - "Balanced perspective"
    label_colors:
      "Too generic/vague": "#ef4444"
      "Excessive superlatives": "#f97316"
      "Narrative/story-like": "#eab308"
      "Lacks specific details": "#f59e0b"
      "Focuses on reviewer not product": "#dc2626"
      "Specific concrete details": "#22c55e"
      "Mentions negatives honestly": "#10b981"
      "Balanced perspective": "#14b8a6"
    tooltips:
      "Too generic/vague": "Could apply to any hotel, lacks specificity"
      "Excessive superlatives": "Too many 'best', 'amazing', 'perfect' claims"
      "Narrative/story-like": "Focuses on the story of their trip rather than the hotel"
      "Lacks specific details": "No mention of specific rooms, features, or experiences"
      "Focuses on reviewer not product": "More about why they traveled than the hotel itself"
      "Specific concrete details": "Mentions specific features, room numbers, staff names"
      "Mentions negatives honestly": "Acknowledges some downsides or imperfections"
      "Balanced perspective": "Neither overly positive nor negative"
    min_selections: 0
    max_selections: 8

  # Step 3: Confidence
  - annotation_type: likert
    name: confidence
    description: "How confident are you in your classification?"
    min_value: 1
    max_value: 5
    labels:
      1: "Just guessing"
      2: "Slightly confident"
      3: "Moderately confident"
      4: "Confident"
      5: "Very confident"

allow_all_users: true
instances_per_annotator: 100
annotation_per_instance: 3
allow_skip: true
skip_reason_required: false