After collecting customer feedback, Clarabridge's sentiment and text analytics engine extracts linguistic content, categorizes and assigns sentiment scores to distinguish the who, what, how, and why of any customer experience. Our proprietary Natural Language Processing (NLP) Engine understands the syntax and context of all the elements of the text—the parts of speech and entities, facts and the linguistic clauses and relationships. This exhaustive approach to NLP creates a foundation to ensure the resulting analysis is comprehensive and accurate. Clarabridge also includes a proprietary language pack framework to extend the NLP capabilities to foreign languages, so customers have one tool to extract insights from any international feedback channel.
Understand Syntax and Context
Behind the scenes, Clarabridge linguistically parses each sentence of each input verbatim (e.g.: customer feedback, call center notes) automatically through a variety of rules-based and statistical-based algorithms. After Clarabridge’s NLP algorithm parses the source text, it can be further tagged and organized by categorization, sentiment calculation, clustering, and structured data mapping algorithms.
Clarabridge offers the flexibility to fine-tune its text analysis platform. For example, the core NLP can be tuned with the addition of customer-specific or industry dictionaries. Clarabridge also provides accuracy audit capabilities for categorization and sentiment to allow users to precisely measure the recall and precision of their categorization and sentiment results.
Clarabridge Natural Language Processing features include:
-
Precise extraction
-
Intelligent and configurable spelling correction.
-
Ability to add customer-specific or industry based dictionaries.
-
Sophisticated features like “anaphora” resolution, which enables the engine to resolve pronouns such as “I” and “it” back to their proper nouns so that issues can be attributed accurately to customers and products.
-
Negation detection, a process during which a negating word (such as ‘not’) inverts the evaluative value of an affective word (for example, “not good” is similar to saying “bad”).
-
Tokenization
-
Block detection
-
Clitics normalization
-
Rule based/machine learning named entity recognition
-
Special entity recognition
-
Unknown word detection using morphological attributes
-
Substitution and filtering
-
Speech part detection and disambiguation
-
Normalization of tokens
-
Misspelling detection
-
Shallow and full syntax parsing
-
Semantic tree generation
-
Clause detection