aws-samples / imdb-conversational-search-blog

IMDb Conversational Search App

This repo contains the code to launch a chatbot app that can help users learn more about the movies in IMDb database through conversation search powered by LLMs and LangChain.

Dataset used: MovieLens + IMDb Mojo Dataset

VectorDB: OpenSearch with 100K records

LLM: Anthropic Claude from Amazon Bedrock, Jurassic Jumbo Instruct, Text2Text, Flan T5 XXL

LLM Framework: LangChain

UI Framework: Streamlit

Features

1. Users have a choice of LLM to select for the conversational search bot:

2. There are two task types that the user can choose from: (Search) or (Search and Chat). First, the LLM searches for movies based on the user query, and then if the chat option is provided, a chat feature is provided to the user.

The UI provides a set of default questions for the search use case as shown below:

These default as well as user inputted search questions can be grouped as either:

• Exact match: searching for movies based on location, actor, plot, rating, directors, etc..
• Semantic match: searching for movies that are similar to others

Currently, the chat bot supports 5 questions at a time to not break context length

Instructions to Run Demo

Set up

1. Run sh setup.sh to install dependencies.
2. Process IMDb dataset as per notebooks/create_datasets and notebooks/IMDB_Dataset_Preprocessing.
3. Create embeddings for IMDb metadata as per notebooks/embedding_generation.
4. Create OpenSearch cluster in the console. We used the cluster configuration below.
   • instance type = r5.large.search
   • EBS volume size = 30 GiB
   • EBS volume type = General Purpose (SSD) - gp2
   • Number of Availability Zones = 1
   • Ensure the cluster is in closed and private VPC. IMDb is a paid proprietary dataset that SHOULD NOT be exposed to external parties.
5. Create OpenSearch index (exact and semantic match)by running python src/index_creation.py. Make sure to configure correct argument values for file paths and index name in the file.

Bedrock Setup: Optional

In order to follow these instructions, you need access to the private beta Bedrock.

In terminal:

1. sh bedrock_install.sh

Test Streamlit Chat App

1. From folder streamlit run streamlit run chat.py
2. In another terminal run bash run.sh to get the local SM Studio app URL

License

This sample code is licensed under the MIT-0 License. See the LICENSE file.