Analyze Text with Azure AI Language Service

Azure AI Text Analytics SDK for Python is a client library for Azure AI Language service NLP capabilities: sentiment, entities, key phrases, PII, language detection, and healthcare text analytics, authenticated via AzureKeyCredential or the recommended DefaultAzureCredential.

analyze_sentiment returns positive/negative/neutral confidence scores per document, and with show_opinion_mining=True breaks sentiment down by aspect - a sentence's mined opinions name a specific target (e.g. "food") and its own sentiment, plus the individual assessment phrases behind it. recognize_entities returns each entity's text, category, subcategory, and confidence score. recognize_pii_entities returns both a redacted version of the text and the specific PII entities found (with category), for detecting things like SSNs and emails. extract_key_phrases returns a flat list of key phrases per document, and detect_language returns the primary language's name, ISO 639-1 code, and confidence. Healthcare text analytics is a long-running operation (begin_analyze_healthcare_entities) that extracts clinical entities with a normalized form and links to external data sources like UMLS. Multiple analyses can run together in one batch via begin_analyze_actions with a list of action objects (RecognizeEntitiesAction, ExtractKeyPhrasesAction, AnalyzeSentimentAction), returning results tagged by kind so you can dispatch on EntityRecognition/KeyPhraseExtraction/SentimentAnalysis. An async client mirrors the sync one for high-throughput use.

Use it when you need NLP analysis of text - sentiment (including aspect-based opinion mining), entities, PII redaction, key phrases, language detection, or clinical entity extraction - rather than building these capabilities yourself. Batch multiple analyses together with begin_analyze_actions rather than calling each method separately when you need more than one kind of analysis on the same documents, since it's explicitly the SDK's batching mechanism for that case.

Input is a list of text documents (up to 10 per request for batch operations), optionally with a specified language to improve accuracy. Output is per-document, per-operation results - sentiment scores and opinions, tagged entities, redacted text, key phrases, detected language, or healthcare entities - and each result must be checked for is_error since a batch call can return per-document errors alongside successes. Install it with:

pip install azure-ai-textanalytics

Its own best practices: batch documents together (up to 10 per request) rather than calling one at a time, enable opinion mining for detailed aspect-based sentiment, use the async client for high-throughput scenarios, always check for per-document errors in the results list, specify the document language when known to improve accuracy, and use a context manager or close the client explicitly.

It's the azure-ai-textanalytics client against the Azure AI Language service, with entity linking resolving against Wikipedia and healthcare entity linking resolving against sources like UMLS.

Python developers building NLP-powered features - sentiment analysis, entity/PII extraction, key phrases, language detection, or clinical text analysis - who want a managed Azure service rather than running their own NLP models.