In this post, you will learn how you can use Amazon Nova Foundation Models in Amazon Bedrock to apply generative AI techniques for both business protection and enhancement. You can identify obvious and disguised attempts at direct contact while gaining valuable insights into customer sentiment and service improvement opportunities.

Using regular expressions (regex) may be the initial solution that comes to mind since it excels at pattern matching and text manipulation, offering a powerful and concise way to search, validate, and transform text data. Regex does well with structured patterns like email addresses, phone numbers, and dates. Contact information follows predictable patterns. Phone numbers use the XXX-XXX-XXXX format, while email addresses follow name@company.com. Regular expressions help identify these patterns in text. For US phone numbers, the regex pattern \d{3}-\d{3}-\d{4} matches three digits, a hyphen, three more digits, another hyphen, and four digits.

However, regex shows significant limitations when dealing with modern text complexities like HTML parsing (because of nested structures and variations in markup), emoji recognition (because of Unicode complexities and variations in emoji representations across services), and evolving patterns like social media handles or changing URLs. Regex falls short when people purposefully conceal contact information by using ever-changing deceptive tactics. For example, a message reading “Congratulations. Here are some more details 555inches 555inches 5555inches” is clearly an attempt to mask a phone number as measurements. Since the pattern is known, a sophisticated regex pattern like (\d+)inches\s+(\d+)inches\s+(\d+)inches can effectively uncover phone numbers concealed by using “inches” as a decoy unit of measurement. But what if the pattern is not as simple as using known measurement identifiers? The challenge of detecting hidden contact information extends beyond simple pattern matching and needs to be more dynamic. Evasion techniques are ever changing such as replacing numbers with words, using alternative units, varying delimiters, and combining leetspeak with emojis. Traditional regex patterns struggle with spelled-out numbers, creative symbol usage (writing “@” as “at”), context-dependent company references, and complex leetspeak combinations.

Regex is particularly inadequate for advanced text analysis needs such as sentiment detection, context understanding, or identifying user actions and intentions in text. For instance, while regex can find specific words, it cannot understand the emotional tone or determine if a user’s message requires follow-up action. When patterns become complex or require frequent updates, regex maintenance becomes challenging and error-prone, often leading to brittle solutions that break when text formats evolve. For these scenarios, consider using generative AI solutions such as Amazon Bedrock. Amazon Bedrock provides sophisticated language models that understand context, parse complex structures, and adapt to evolving text patterns without constant manual updates.

Amazon Bedrock is a fully managed, serverless service offering a variety of high-performing AI foundation models from leading companies. You can use Amazon Bedrock to experiment with, customize, and integrate generative AI capabilities into your applications using familiar AWS services. Amazon Bedrock also provides a playground feature on the AWS console to test prompts and multiple LLMs.