Transforming Lex Bot Utterances into Actionable Data with AWS DataBrew

AWS Lex bot offers benefits like enhanced user engagement, scalability, and the ability to integrate with a broad ecosystem of AWS services. It allows businesses to deliver a superior customer experience, with engaging, natural, and interactive dialogues. While AWS DataBrew significantly simplifies the tasks of data exploration and transformation. It eliminates the need for coding, accelerates the process of preparing data for analysis, and makes the data transformation task accessible with minimum management overhead.

By integrating these two tools, we can streamline user utterance data from Lex bots using DataBrew, thus enhancing bot performance and generating valuable data for a range of business needs. From robust analytics reporting to retraining machine learning models, the transformed data serves as a key enabler for informed decision-making and operational optimization.

In this article, we will demonstrate to streamline the data transformation of AWS Lex bot utterances into a resource of actionable data, thereby fueling informed decision-making and enhancing business performance.

Unlocking Business Potential with AWS Lex: Features and Benefits

As businesses increasingly embrace digital transformation, an often overlooked yet profoundly impactful element is the evolution of customer interaction mechanisms. In this realm, the Amazon Web Services (AWS) Lex bot is a game-changing tool. Providing capabilities for building conversational interfaces, AWS Lex harnesses the power of advanced deep learning functionalities of automatic speech recognition (ASR) for converting speech to text, and natural language understanding (NLU) to recognize the intent of the text.

Key Features of AWS Lex

• Conversational Interfaces: AWS Lex allows you to create sophisticated, natural language conversational bots ("chatbots"), which can recognize and respond to user inputs in a human-like manner. These interfaces can be text or voice-based, offering flexibility in deployment across multiple platforms.
• Deep Learning Technologies: AWS Lex leverages the same deep learning technologies as Alexa, Amazon's virtual assistant. It uses automatic speech recognition (ASR) to convert spoken language into written text, and natural language understanding (NLU) to comprehend the user's intent from the text.
• Integration Capabilities: AWS Lex can be seamlessly integrated with various AWS services like AWS Lambda for executing business logic, Amazon Cognito for authentication, and Amazon CloudWatch for logging and monitoring.
• Multi-Platform Compatibility: You can deploy your Lex bot on multiple platforms including mobile devices, web applications, and chat services such as Facebook Messenger, Slack, and Twilio.
• Scalability and Security: AWS Lex scales automatically, so you don't have to worry about infrastructure provisioning, management, or security as your usage grows.

Advantages of AWS Lex

AWS Lex offers a highly customizable, scalable, and user-friendly platform for designing and deploying chatbots. By transforming the way businesses interact with their users, AWS Lex is reshaping the customer service landscape, paving the way for more efficient, engaging, and personalized customer experiences.

• Enhanced Customer Experience: By providing immediate, automated responses to customer inquiries, AWS Lex enhances user engagement and satisfaction. Conversational interfaces offer a more interactive and personalized user experience, leading to higher customer retention.
• Operational Efficiency: AWS Lex can handle routine tasks and inquiries, freeing up your human resources to focus on more complex and value-adding tasks. This can significantly improve your organization's operational efficiency and reduce costs.
• Insightful Analytics: AWS Lex provides built-in analytics to track usage metrics, missed utterances, and conversation paths. These insights can help you continually refine and improve your bot's performance, leading to more effective customer interactions.
• Easy to Use: With AWS Lex, there's no need to have expertise in machine learning. The service provides an easy-to-use console for creating your own chatbot, allowing you to focus on designing the best conversational experiences for your users.

Use Utterance recognition data to improve bot performance

Utterance recognition is a crucial part of the natural language understanding (NLU) process that Amazon Web Services' Lex bot employs to comprehend user inputs. Essentially, an utterance is any input that a user might say or type to interact with the Lex bot. This could be a simple command, a question, or a complex sentence. The main goal of utterance recognition is to interpret these inputs accurately, extract the intended meaning, and trigger the appropriate response from the bot.

When designing a Lex bot, defining a comprehensive set of utterances is key. These should cover a wide range of potential inputs that users might enter, essentially predicting the various ways users might phrase their requests or inquiries. For instance, if your bot is designed to book appointments, your utterances could include phrases like "I want to schedule a meeting," "Book an appointment," or "Can I arrange a meeting?"

Detected utterance vs missed utterance

• Detected Utterance: A detected utterance is an input that the bot successfully recognizes and matches with one of its predefined intents. In other words, the bot understands the input and knows how to respond based on the training it has received. For instance, if a user says "Book an appointment for me," and the bot has been trained to recognize this phrase as an intent to schedule an appointment, this is a detected utterance.
• Missed Utterance: A missed utterance, on the other hand, is an input that the bot fails to recognize or match with any of its predefined intents. This can happen if the user says something in a way that the bot has not been trained to understand. For example, if a user says "I need to pencil in a meeting," but the bot has not been trained to understand this phrase as an intent to schedule an appointment, this would be a missed utterance.

AWS Lex provides a missed utterances report that shows utterances that were not recognized by the bot. This report can be invaluable for improving the bot's performance. By identifying the phrases that users commonly say but the bot doesn't understand, you can add these as new utterances to the bot's existing intents, or create new intents as necessary. This iterative process helps to continually enhance the bot's ability to understand and respond to user inputs.

Enhancing Data Transformation with AWS DataBrew

AWS DataBrew is a powerful tool that specializes in data transformation. This fully managed, visual data preparation solution is engineered to clean and normalize data for analysis and machine learning. Its standout feature is the ability to transform data without the need to write any code, making it accessible to users with varying degrees of technical proficiency.

When applied to Amazon Lex bot's utterance data, AWS DataBrew becomes instrumental in enhancing the quality of insights. For instance, it can transform raw conversational data into structured, usable datasets, improving the quality of insights drawn from your bot's interactions. The solution offers over 250 pre-built transformations that can cleanse, combine, and reshape data with ease, thus streamlining the process of data transformation, reducing manual errors, and increasing productivity.

Unlocking Benefits with AWS DataBrew

Utilizing AWS DataBrew to handle Lex bot's utterance data offers a wealth of benefits:

• Efficiency: AWS DataBrew is designed to enhance operational efficiency by reducing the time spent on data preparation, which typically consumes up to 80% of data analysts' time. By automating this process, DataBrew frees up time for more value-added tasks such as data analysis and interpretation.
• Collaboration: The solution fosters a collaborative environment. It allows data professionals to share their transformation recipes with others, promoting knowledge sharing within the team.
• Accessibility: The visual interface of DataBrew makes it easy for team members to understand and participate in the data transformation process, fostering a more inclusive data analysis culture.
• Scalability: AWS DataBrew scales according to the size of datasets, making it ideal for businesses of all sizes.

Lex utterance analytics data ETL with Lambda and DataBrew

Architecture

In the Architecture of our example, the raw interaction data from users were collected by Amazon Lex and then transformed by backend services, and eventually becomes training data for advanced machine learning or reporting data for dashboard analytics.

In this workflow, the user interacts with the Lex bot by providing inputs. The Lex bot then collects the users’ inputs, categorizing the input data as detected utterance or missed utterance, both of which combined into the utterance recognition data. The utterance recognition data is then queried and transferred to a raw S3 bucket for initial data storage. After that, the Glue DataBrew job is triggered to process the raw data and store the transformed data in a separate S3 bucket as data source for downstream training data for Machine Learning, Lex bot improvement or business analytics purposes.

Lex Bot set up

As a prerequisite for our implementation example, we have configured a Lex bot “AutoInsuranceBot” with pre-defined bot intents.

The simplest way to create the Lex bot is using AWS Lex via the AWS Management Console: First define the intents and utterances that your bot will understand, and configure any necessary slots and prompts. In this example, we used a pre-built bot template in Lex with default intents for AutoInsurance assistant. At the end of creating the Lex bot, the console will direct to the creation of a pre-configured CloudFormation stack which will help deploy the resources needed for this Lex bot.

More details on the example of the Lex Bot set-up in this post.

Extract raw utterance data with Lambda

In our example, we set up a lambda function to run on-demand invocations to extract the utterance data collected by Lex bot. The “lexv2-models” client in AWS SDK for Python (Boto3) provides a convenient way to query Lex utterance analytics data by using filters and defined time frames. In this function, we first specify our Lex bot’s botId, we also filtered the bot’s “BotAliasId” and its value to identify the latest bot deployment for utterance data collection. Then we listed all the available utterance data from user inputs during 2024/5/14 00:00:00 – 14:00:00. Finally, we put this raw data into the S3 raw bucket in a json format for Glue DataBrew to process.

Streamline data preparation with Glue DataBrew

In this workflow, we choose Glue DataBrew for data transformation because of its compelling advantages in visual data exploration and minimum development overhead with fully-managed code-free environment. In DataBrew, we first create a “dataset” which connects to the raw utterance json data in S3.

After that, we create a DataBrew project that contains the connected DataBrew dataset and a DataBrew recipe that contains procedures for data transformation. In this example, we want to create the recipe from scratch as we explore the structure of the raw dataset and the final schema of the transformed dataset.

After the project is created, it automatically provisions a compute resource and provides a view for the raw dataset’s schema with sample data rows. In this comprehensive visual aid, we can quickly explore the structure of the raw dataset and add steps to recipe from pre-built

In this dataset, we applied multiple steps in the recipe to extract columns such as data related to bot responses and user input. The pre-built data transformation method helped apply these steps directly to the sample data rows for visual results as we develop them, for example, unnest values that was nested as key-value format and mask original user data.

After the recipe is developed, we ran the DataBrew job which processed the entire raw dataset and saved the transformed output in a csv format in the S3 bucket. The cleaned Lex bot utterance data is now prepared and primed for downstream user integration. It can be leveraged in various business logics for:

• Business Reporting: Users can directly utilize the cleaned data with AWS QuickSight for business reporting purposes.
• Improving Bot Performance: The data can be fed back to Lex bot to enhance its performance.
• Advanced Machine Learning: The processed data can be used as input data for building models with AWS SageMaker.
• Fine-tuning Gen-AI Models: The data can also be utilized for fine-tuning Gen-AI models using AWS BedRock.

Conclusion

The integration of AWS Lex bot and AWS DataBrew unveils a robust and technologically advanced approach to manage and optimize user utterance data. The AWS Lex bot, with its superior natural language understanding capabilities, proves invaluable in capturing and interpreting user utterances efficiently. On the other hand, AWS DataBrew emerges as a powerful data preparation tool that aids in cleaning, transforming, and normalizing the data obtained from AWS Lex bot. This ensures that the data is in an optimal format for further analysis or processing. The streamlined workflow resulting from this integration significantly enhances the efficiency of data management, ensuring that user utterances are not just comprehended but also utilized effectively. By leveraging these advanced AWS services, businesses can unlock new levels of performance and derive greater insights from their data, ultimately fostering a more user-centric and data-driven approach.

Reference

AWS Lex v2 ?– User Guide: https://docs.aws.amazon.com/lexv2/latest/dg/what-is.html

AWS Glue DataBrew - Developer Guide:

https://docs.aws.amazon.com/databrew/latest/dg/getting-started.html

Authors

Vinny Wang

Vinny Wang is a Software Engineer at Deloitte Consulting, specializing in Cloud Native Development on the AWS platform holding all 13 active AWS certifications. With extensive experience in developing end-to-end data driven machine learning solutions across industries, he excels at creating effective solutions in all stages of cloud infrastructure and machine learning project cycles.

Gaurav Kumar

Gaurav Kumar is a Senior Consultant at Deloitte Consulting, specializing in cloud infrastructure architecture design and cloud-native development. With 11 years of technical and consulting experience, he has worked as an architect on several large-scale implementations and has assisted various clients in adopting best practices in infrastructure solutions such as automation, infrastructure as code, and migration.

Sponsor

Gowtham Ramu

Gowtham Ramu is a Specialist Leader in Cloud Engineering at Deloitte Consulting, bringing over 24 years of experience in crafting solutions that span various industries. His expertise lies in developing Cloud Native solutions at scale, with a strong focus on data engineering. With more than 10 AWS certifications, he instills in his team a commitment to continuous learning as a fundamental principle.