Skip to content

Explore Data

To access and explore the ingested data:

  1. Go to Azure Portal

  2. Locate the Resource Group where the solution is deployed

  3. Click on the Storage Account associated with the solution image

  4. Navigate to Containers section image

  5. Open the container named data

You’ll see two folders:

  • audiodata → Contains uploaded call recordings image

call_transcripts → Stores the transcript text used for AI processing image

Data Flow

Data Processing

When you deploy this solution, the Process Data script gathers call transcripts and audio data, which are analyzed using Azure AI Content Understanding to extract essential details such as conversation IDs, call content, summaries, sentiments, and more. This processed data is subsequently stored in an Azure SQL database for ongoing analysis and easy retrieval. The script also generates text embeddings, which are uploaded to an Azure AI Search index for future retrieval. Additional details about the processed data :

Text Analysis

  • Sentiment Analysis: Determines the overall sentiment of the conversation (Positive or Negative).
  • Topic Mining: Identifies the main topic of the conversation (e.g., Billing Issues, Device Troubleshooting).
  • Key Phrase Extraction: Highlights important phrases for quick insights.
  • Complaint Identification: Extracts specific complaints raised by the customer.

Structuring the Data

  • The analyzed data is structured into JSON format for easy querying and visualization.

Analyzer Workflow

Sentiment Analysis

  • Uses Natural Language Processing (NLP) to classify the sentiment as Positive or Negative.
  • Example: "Thank you for your help" → Positive sentiment.

Topic Mining

  • Identifies the main topic of the conversation using keyword matching and clustering.
  • Example: Keywords like "billing," "charges," and "refund" → Topic: Billing Issues.

Key Phrase Extraction

  • Extracts important phrases using NLP techniques like Named Entity Recognition (NER).
  • Example: "slow Internet speed," "paperless billing," "factory reset."

Complaint Identification

  • Searches for explicit complaints in the conversation.
  • Example: "My bill was $50 higher than usual" → Complaint: Higher bill.