Usage

To enable semantic search on a text field, set the attribute below:

When searching for records:

Technical Implementation

The module works by:

1. Creating pickle files that store document representations of text fields.
2. When a search runs on a field with semantic_search=True it:
3. • Uses stored document representations
   • Applies semantic search algorithms to find relevant matches
   • Ranks results using multiple scoring factors
   • Returns matching record IDs

The Bista Text Semantic Search is an advanced document retrieval system for Odoo, built on the Haystack framework. It integrates traditional IR with modern NLP to deliver precise search results across project documentation. The system follows a three-tier architecture:

• Data Layer: Documents are serialized with pickle, enriched with metadata such as project names and content hashes.
• Processing Layer: Uses Haystack’s InMemoryDocumentStore with the BM25 algorithm for fast in-memory retrieval.
• Application Layer: Adds fuzzy query correction (RapidFuzz), semantic validation (WordFreq), and composite scoring mechanisms.

Queries undergo Zipf frequency analysis to filter meaningful terms. High-scoring queries proceed to retrieval, while suspected typos trigger fuzzy correction. BM25 ranks documents using term frequency saturation, length normalization, and inverse document frequency. Results are reranked through multi-factor scoring: calibrated BM25, query coverage, exact match boosts, and term density. When retrieval fails, intelligent fallbacks like prefix matching are applied. Final results are sorted by composite score and returned as ranked dictionaries containing project names, normalized relevance scores (0–100), and full content for immediate display.