Content Safety Filter

Content Safety Filter

In a live commerce platform, buyers and sellers interact in real-time chat during auctions. To prevent fraud and abuse, incoming chat messages must be filtered to mask prohibited terms and phrases before being broadcasted.

Write a function filter_content(message: str, banned_phrases: list[str]) -> str that detects all occurrences of prohibited phrases in message and replaces every alphanumeric character in the matched words with an asterisk (*). Non-alphanumeric characters (punctuation, symbols, whitespace) must be preserved in their original form and positions.

Rules

• Case-Insensitive: Matching between message and banned_phrases is case-insensitive.
• Word Boundaries: Matches must align with word boundaries. A word is defined as a maximal contiguous sequence of alphanumeric characters ([a-zA-Z0-9]). Punctuation and whitespace act as delimiters.
• Multi-Word Phrases: A banned phrase may contain multiple words separated by single spaces. In the message, words in a multi-word phrase can be separated by any sequence of non-alphanumeric characters (whitespace or punctuation).
• Preservation: Punctuation and spacing must remain intact; only the alphanumeric characters of matched words are masked.
• Overlaps: If matched phrases overlap, all characters belonging to any matched phrase must be masked.

Function Signature

def filter_content(message: str, banned_phrases: list[str]) -> str:
    pass

Constraints

• 1 <= len(message) <= 10^5
• 1 <= len(banned_phrases) <= 5 * 10^4
• Total words across all banned phrases <= 10^5
• banned_phrases contains only alphanumeric words separated by single spaces.
• Target time complexity: O(N + M) where N is the length of message and M is the total character length of banned_phrases.

Examples

Example 1:

message = "Check out this fake giveaway! Don't buy."
banned_phrases = ["fake giveaway", "buy"]
output = filter_content(message, banned_phrases)
# output: "Check out this **** ********! Don't ***."

Explanation: "fake giveaway" matches the two words "fake" and "giveaway". "buy" matches "buy". All alphanumeric characters are replaced with *, preserving whitespace and punctuation.

Example 2:

message = "Scam-alert: Free coins at url!"
banned_phrases = ["scam alert", "free coins"]
output = filter_content(message, banned_phrases)
# output: "****-*****: **** ***** at url!"

Explanation: In "Scam-alert", the hyphen acts as a non-alphanumeric separator, matching the phrase "scam alert".

Edge Cases

• Partial word matches must not trigger a replacement (e.g., "buy" should not match inside "buyer").
• Consecutive punctuation or whitespace between words should not disrupt multi-word phrase matching.