{"id":434,"date":"2020-10-21T22:47:09","date_gmt":"2020-10-21T22:47:09","guid":{"rendered":"https:\/\/machine-learning.webcloning.com\/2020\/10\/21\/time-series-forecasting-using-unstructured-data-with-amazon-forecast-and-the-amazon-sagemaker-neural-topic-model\/"},"modified":"2020-10-21T22:47:09","modified_gmt":"2020-10-21T22:47:09","slug":"time-series-forecasting-using-unstructured-data-with-amazon-forecast-and-the-amazon-sagemaker-neural-topic-model","status":"publish","type":"post","link":"https:\/\/salarydistribution.com\/machine-learning\/2020\/10\/21\/time-series-forecasting-using-unstructured-data-with-amazon-forecast-and-the-amazon-sagemaker-neural-topic-model\/","title":{"rendered":"Time series forecasting using unstructured data with Amazon Forecast and the Amazon SageMaker Neural Topic Model"},"content":{"rendered":"
\n

As the volume of unstructured data such as text and voice continues to grow, businesses are increasingly looking for ways to incorporate this data into their time series predictive modeling workflows. One example use case is transcribing calls from call centers to forecast call handle times and improve call volume forecasting. In the retail or media industry, companies are interested in using related information about products or content to forecast popularity of existing or new products or content from unstructured information such as product type, description, audience reviews, or social media feeds. However, combining this unstructured data with time series is challenging because most traditional time series models require numerical inputs for forecasting. In this post, we describe how you can combine Amazon SageMaker<\/a> with Amazon Forecast<\/a> to include unstructured text data into your time series use cases.<\/p>\n

Solution overview<\/h2>\n

For our use case, we predict the popularity of news articles based on their topics looking forward over a 15 day horizon. You first download and preprocess the data and then run the NTM algorithm to generate topic vectors. After generating the topic vectors, you save them and use these vectors as a related time series to create the forecast.<\/p>\n

The following diagram illustrates the architecture of this solution.<\/p>\n

\"\"<\/p>\n

AWS services<\/h2>\n

Forecast is a fully managed service that uses machine learning (ML) to generate highly accurate forecasts without requiring any prior ML experience. Forecast is applicable in a wide variety of use cases, including energy demand forecasting, estimating product demand, workforce planning, and computing cloud infrastructure usage.<\/p>\n

With Forecast, there are no servers to provision or ML models to build manually. Additionally, you only pay for what you use, and there is no minimum fee or upfront commitment. To use Forecast, you only need to provide historical data for what you want to forecast, and, optionally, any related data that you believe may impact your forecasts. This related data may include time-varying data (such as price, events, and weather) and categorical data (such as color, genre, or region). The service automatically trains and deploys ML models based on your data and provides you with a custom API to retrieve forecasts.<\/p>\n

Amazon SageMaker is a fully managed service that provides every developer and data scientist with the ability to build, train, and deploy ML models quickly. The Neural Topic Model<\/a> (NTM) algorithm is an unsupervised learning algorithm that can organize a collection of documents into topics that contain word groupings based on their statistical distribution. For example, documents that contain frequent occurrences of words such as \u201cbike,\u201d \u201ccar,\u201d \u201ctrain,\u201d \u201cmileage,\u201d and \u201cspeed\u201d are likely to share a topic on \u201ctransportation.\u201d You can use topic modeling to classify or summarize documents based on the topics detected. You can also use it to retrieve information and recommend content based on topic similarities.<\/p>\n

The derived topics that NTM learns are characterized as a latent representation because they are inferred from the observed word distributions in the collection. The semantics of topics are usually inferred by examining the top ranking words they contain. Because the method is unsupervised, only the number of topics, not the topics themselves, are pre-specified. In addition, the topics aren\u2019t guaranteed to align with how a human might naturally categorize documents. NTM is one of the built-in algorithms you can train and deploy using Amazon SageMaker.<\/p>\n

Prerequisites<\/h2>\n

To follow along with this post, you must create the following:<\/p>\n

To create the aforementioned resources and clone the forecast-samples<\/code> GitHub repo into the notebook instance, launch the following AWS CloudFormation<\/a> stack:<\/p>\n

\"\"<\/a><\/p>\n

In the Parameters <\/strong>section, enter unique names for your S3 bucket and notebook and leave all other settings at their default.<\/p>\n

\"\"<\/p>\n

When the CloudFormation script is complete, you can view the created resources on the Resources <\/strong>tab of the stack.<\/p>\n

\"\"<\/p>\n

Navigate to Sagemaker and open the notebook instance created from the CloudFormation template. Open Jupyter and continue to the \/notebooks\/blog_materials\/Time_Series_Forecasting_with_Unstructured_Data_and_Amazon_SageMaker_Neural_Topic_Model\/<\/code> folder and start working your way through the notebooks.<\/p>\n

Creating the resources manually<\/h3>\n

For the sake of completeness, we explain in detail the steps necessary to create the resources that the CloudFormation script creates automatically.<\/p>\n

    \n
  1. Create an IAM role that can do the following:\n
      \n
    1. Has permission to access Forecast and Amazon S3 to store the training and test datasets.<\/li>\n
    2. Has an attached trust policy to give Amazon SageMaker permission to assume the role.<\/li>\n
    3. Allows Forecast to access Amazon S3 to pull the stored datasets into Forecast.<\/li>\n<\/ol>\n<\/li>\n<\/ol>\n

      For more information, see Set Up Permissions for Amazon Forecast<\/a>.<\/p>\n

        \n
      1. \nCreate an Amazon SageMaker notebook instance<\/a>.<\/li>\n
      2. Attach the IAM role you created for Amazon SageMaker to this notebook instance.<\/li>\n
      3. \nCreate an S3 bucket<\/a> to store the outputs of your human workflow.<\/li>\n
      4. Copy the ARN of the bucket to use in the accompanying Jupyter notebook.<\/li>\n<\/ol>\n

        \u00a0<\/strong>This project consists of three notebooks, available in the GitHub repo<\/a>. They cover the following:<\/p>\n