Importing conversation transcripts

Importing conversation transcripts is a three-step process:

1. Prepare the transcripts for importing by converting them to the correct format. If you are using Contact Lens for Amazon Connect the transcripts are already in the correct format.

2. Upload the transcripts to an Amazon S3 bucket. If you are using Contact Lens, your transcripts are already in an S3 bucket.

3. Analyze the transcripts using the Amazon Lex V2 console or API operations. The time that it takes to complete training depends on the volume of transcripts and the complexity of the conversation. Typically, 500 lines of transcripts are analyzed every minute.

Each of these steps is described in the following sections.

Importing transcripts from Contact Lens for Amazon Connect

The Amazon Lex V2 automated chatbot designer is compatible with Contact Lens transcript files. To use Contact Lens transcript files, you must turn on Contact Lens and note the location of its output files.

To export transcripts from Contact Lens

1. Turn on Contact Lens in your Amazon Connect instance. For instructions, see Enable Contact Lens for Amazon Connect in the Amazon Connect administrator guide.

2. Note the location of the S3 bucket that Amazon Connect is using for your instance. To see the location, open the Data storage page in the Amazon Connect console. For instructions, see Update instance settings in the Amazon Connect administrator guide.

After you have turned on Contact Lens and noted the location of your transcript files, go to Analyze your transcripts using Amazon Lex V2 console for instructions to import and analyze your transcripts.

Prepare transcripts

Prepare your transcripts by creating transcript files.

• Create one transcript file per conversation listing the interaction between the parties. Each interaction in the conversation can span multiple lines. You can provide both redacted and non-redacted versions of the conversation.

• The file must be in the JSON format specified in Input transcript format.

• You must provide at least 1,000 conversational turns. To improve the discovery of your intents and slot types, you should provide around 10,000 or more conversational turns. The automated chatbot designer will only process the first 700,000 turns.

• There is no limit to the number of transcript files that you can upload, nor is there a file size restriction.

If you plan to filter the transcripts that you import by date, the files must be in the following directory structure:

<path or bucket root>
   --> yyyy
      --> mm
         --> dd
            --> transcript files

The transcript file must contain the date in the format "yyyy-mm-dd" somewhere in the file name.

To export transcripts from other contact center applications

1. Use your contact center application's tools to export conversations. The conversation must contain at least the information specified in Input transcript format.

2. Transform the transcripts produced by your contact center application to the format described in Input transcript format. You are responsible for performing the transformation.

We provide three scripts for preparing transcripts. They are:

• A script to combine Contact Lens transcripts with Amazon Lex V2 conversation logs. Contact Lens transcripts don't include parts of Amazon Connect conversations that interact with Amazon Lex V2 bots. The script requires conversation logs to be turned on for Amazon Lex V2, and appropriate permissions to query conversation log CloudWatch Logs and Contact Lens S3 buckets.

• A script to transform Amazon Transcribe call analytics to the Amazon Lex V2 input format.

• A script to transform Amazon Connect chat transcripts to the Amazon Lex V2 input format.

You can download the scripts from this GitHub repository: https://github.com/aws-samples/amazon-lex-bot-recommendation-integration .

Upload your transcripts to an S3 bucket

If you are using Contact Lens, your transcript files are already contained in an S3 bucket. For the location and file names of your transcript files, see Example Contact Lens output files in the Amazon Connect administrator guide.

If you are using another contact center application and you have not set up an S3 bucket for your transcript files, follow this procedure. Otherwise, if you have an existing S3 bucket, after logging in to the Amazon S3 console, follow this procedure starting with step 5.

To upload files to an S3 bucket

1. Sign in to the AWS Management Console and open the Amazon S3 console at https://console.aws.amazon.com/s3/.

2. Choose Create bucket.

3. Give the bucket a name and choose a Region. The Region must be the same one that you use for Amazon Lex V2. Set the other options as required for your use case.

4. Choose Create bucket.

5. From the list of buckets, choose an existing bucket or the bucket that you just created

6. Choose Upload.

7. Add the transcript files that you want to upload.

8. Choose Upload.

Analyze your transcripts using Amazon Lex V2 console

You can only use automated bot design in an empty language. You can add a new language to an existing bot, or create a new bot.

To create a new language in a new bot

1. Sign in to the AWS Management Console and open the Amazon Lex console at https://console.aws.amazon.com/lex/.

2. Choose Create bot

3. Choose Start with Automated Chatbot Designer. Fill out the information to create your new bot.

4. Choose Next

5. In Add language to bot fill out the information for the language.

6. In the Transcript file location on S3 section, choose the S3 bucket that contains your transcript files and the local path to the files if necessary.

7. You can optionally choose the following:

   • A AWS KMS key to encrypt the transcript data during processing. If you don't select a key, a service AWS KMS key is used.

   • To filter the transcripts to a specific date range. If you choose to filter the transcripts, they must be in the correct folder structure. For more information, see Prepare transcripts.

8. Choose Done.

Wait for Amazon Lex V2 to process the transcript. You see a completion message when the analysis is complete.

How to stop analyzing your transcript

In case you need to stop the analysis of the transcripts you have uploaded, you can stop a running BotRecommendation job, which has a BotRecommendationStatus status as processing. You can click on the Stop processing button present on the banner after submitting a job from the console or by using CLI SDK for the StopBotRecommendation API. For more information, see StopBotRecommendation

After calling the StopBotRecommendation, the internal BotRecommendationStatus is set to Stopping and you are not charged. To make sure the job has stopped, you can call the DescribeBotRecommendation API and verify that the BotRecommendationStatus is Stopped. This usually takes 3-4 minutes.

You are not charged for the processing after the StopBotRecommendation API is called.