Tisane is a powerful Natural Language Understanding (NLU) platform, designed from scratch for social media / user-generated text: ungrammatical, possibly obfuscated (e.g. algospeak), and heavy on slang and errors. Tisane's main applications are content moderation and law enforcement/national security/intelligence.
All features are available across all supported languages. All output is provided in one call.
Tisane offers sophisticated capabilities for abuse detection, sentiment analysis, and entity recognition:
Detect problematic content for content moderation and investigations:
- Insults, cyberbullying, personal attacks
- Hate speech
- Profanity
- Sexual solicitation
- Adult only references
- Criminal activity (trade in restricted items like drugs, firearms, etc.)
- Threats
- Allegations (whether libel or not)
- Suicidal ideation
- Solicitation of contact outside the current media or platform
- Graphic depictions of violence
- Forceful assertions of hierarchy (e.g. workplace bullying)
- Data leaks (e.g. PII exfiltration, exposure of credentials)
- Spam
Detect named entities and find relations between them:
- people
- organizations
- social roles (titles, occupations, etc.)
- places
- phone numbers
- credit card numbers
- dates
- amounts of money
- items of interest in investigations (drugs, weapons, vehicles)
- cryptocurrency addresses
- more
Detect topics (Subjects/Themes/Domains)
Sentiment analysis (positive, negative, or mixed), including
- sentiment expressed towards specific aspects or entities within the text. (Aspect-based sentiment analysis)
Tisane also provides access to linguistic fundamentals:
- Inteligent sentence splitting
- Provides original sentences and their corrected versions (if applicable).
- Uses non-breaks to avoid false positives (e.g.
EE. UU.orwww.google.com)
- Lexical chunking
- Tokenizes sentences by morpheme
- Supports all types of segmentation:
- languages using white spaces
- languages not using white spaces (e.g. Chinese, Japanese, Thai)
- languages using compounds (German, Dutch, Afrikaans)
- clitics
- multiword expressions
- nondictionary entities (emails, phone numbers, etc.)
- Parse Trees and Phrases
- Offers hierarchical representations of sentence structure, highlighting phrases and their relationships.
- Extracts noun phrases, verbal phrases, prepositional phrases, adjectival phrases, adverbial phrases.
- Words
- Provides detailed information about each word (or lexical chunk), including its lemma, role in the sentence, and grammatical features.
- Sense disambiguation
- Detects the sense for every word in the current context.