Let’s write some code.

We’re going to build a simple semantic search example.

Semantic Search Example: Movie Recommendations

Scenario: Imagine you have a movie database, and you want to build a recommendation system. A user recently watched and loved the movie "Inception" and wants similar movie recommendations.

Traditional Search: A traditional search might look for movies that have the word "Inception" in their titles, descriptions, or reviews. It would return movies like "Inception: The Cobol Job" (a related short film) or any movie where a reviewer mentioned "Inception". This method is very literal and might miss out on movies that have a similar feel or theme but don't explicitly mention "Inception".

Semantic Search: Instead of just looking for the word "Inception", a semantic search engine would understand the deeper themes and vibes of "Inception", such as dream manipulation, layered realities, heist, and mind-bending plots. It would then recommend movies that share these themes, even if they never mention "Inception" or dreams at all.

Recommendations: With semantic search, the user might get recommendations like:

• "The Matrix" - Deals with alternate realities and questioning what's real.
• "Shutter Island" - A mind-bending plot with twists and turns.
• "Memento" - A non-linear narrative that challenges the viewer's perception.
• "Doctor Strange" - Features alternate dimensions and reality manipulation.

Benefits:

1. Relevance: The recommendations are more in line with what the user is likely seeking – movies that "feel" like "Inception", not just movies that mention it.
2. Discoverability: Users discover movies they might never have come across in a traditional search.
3. User Satisfaction: By meeting or even exceeding user expectations, satisfaction and trust in the recommendation system increase.

1. Imports:

import 'dotenv/config'
import { Document } from 'langchain/document'
import { MemoryVectorStore } from 'langchain/vectorstores/memory'
import { OpenAIEmbeddings } from 'langchain/embeddings/openai'

• The code begins by importing necessary modules.
  • dotenv/config: Loads environment variables from a .env file.
  • langchain/document: Imports the Document class, used to structure data for vector indexing.
  • langchain/vectorstores/memory: Imports the MemoryVectorStore, an in-memory storage for vectors (semantic representations of data).
  • langchain/embeddings/openai: Imports the OpenAIEmbeddings, which interfaces with OpenAI to obtain semantic embeddings for data.