Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

fix: update rrf formula #117

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix: update rrf formula #117

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 8 additions & 9 deletions src/server/api/core/rag/retrieval/search-method.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,8 @@
import { and, desc, eq, sql } from 'drizzle-orm'
import { SearchTypeEnum } from '~/model/search-type'
import { db } from '~/server/db'
import * as schema from '~/server/db/migration/schema'
import getEmbeddingsFromContents from '~/server/gateway/openai/embedding'
import { SearchTypeEnum } from '~/model/search-type'

interface RankedResult {
content: string | null
Expand Down Expand Up @@ -198,18 +198,17 @@ async function hybridSearch(
// Combine results
const combinedResults = combineSearchResults(vectorResults, fullTextResults)

const totalResults = vectorResults.length + fullTextResults.length

// Calculate RRF scores
// Assume we search 20 chunk for each search ( vector search and full-text search )
// if vector rank is [10, 41] => An item ranked 10th in vector search but not found in full-text would have ranks
// if full-text rank is [41, 20] => An item ranked 20th in full-text but not found in vector search would have ranks
// Using totalResults + 1 for penalty rank
// formula: RRF = 1 / (k + rank) for each rank
// Assume we have 2 search results: vector search and full-text search
// For each result d, we have 2 ranks: rank d in vector search and rank d in full-text
// If d found in both searches, then RFF score d = 1 / (k + rank d in vector search) + 1 / (k + rank d in full-text search)
// If d found in vector search but not in full-text search, then RFF score d = 1 / (k + rank d in vector search) + 1 / (k + (len(full-text search) + 1))

combinedResults.forEach((result) => {
const ranks = [
result.vectorRank ?? totalResults + 1,
result.textRank ?? totalResults + 1,
result.vectorRank ?? vectorResults.length + 1,
result.textRank ?? fullTextResults.length + 1,
]

result.rrfScore = calculateRRFScore(ranks)
Expand Down