Semantic Movie Search is a simple python app powered by Elasticsearch and txtai.
It takes in a user input, gets the top 50 Elasticsearch results, then runs a txtai semantic similarity function on the top 500 Elasticsearch results, reranks them, and returns the new top 50 to compare.
You need Python installed, and a few dependencies, specifically Streamlit (for running the app), Elasticsearch, and txtai.
This project assumes you have a local Elasticsearch cluster running on port 9200. You can run a local cluster and setup a movies index via docker-compose from the simple-reranker project repo. The only addition you'll need to make to the Elasticsearch cluster is adding the semantic_search_test search template located in the root of the project.
You can add it in Kibana with the following command:
PUT _scripts/semantic_search_test
{
"script": {
"lang": "mustache",
"source": """
{
"query": {
"multi_match": {
"fields": [
"title",
"description",
"genres.name"
],
"query": "{{query_string}}"
}
}
}
"""
}
}
The searching and reranking is not fast. That's because a similarity function is ran against each Elasticsearch result (essentially converting text to vector embeddings on the fly.) This works well for a proof of concept, but ideally, you would convert the data to vectors and add to a new Elasticsearch index, or vector database.