query-understanding

Awesome Search - this is all about the (e-commerce, but not only) search and its awesomeness

Systematic Review Query Visualisation and Understanding Interface

IntentsKB: A Knowledge Base of Entity-Oriented Search Intents - CIKM'18

Towards an Understanding of Entity-Oriented Search Intents - ECIR'18

Presentation and Code for talk at Conferences - MLDS-2020 and DHS-2019

Target Type Identification for Entity-Bearing Queries - SIGIR'17

模块化、可组合的 RAG 检索库 —— 像锻造零件一样构建你的检索管道。Modular, composable RAG retrieval library —— Build your retrieval pipeline like forging parts.

Production-grade search engine with hybrid retrieval (BM25 + FAISS), Learning-to-Rank, neural reranking, and real-time query understanding for scalable information retrieval systems.

A hybrid retrieval-driven desktop engine that indexes Zotero PDFs into a high-fidelity semantic layer, enabling context-aware querying, source-grounded responses, and fine-grained metadata control across complex research corpora.

A new package that processes user queries about why small voting or ranking projects get flagged as spam so easily. It uses natural language processing to understand the input and generates a structur

Curated papers and resources for complex query understanding, e-commerce product retrieval, semantic IDs, and agentic search.

QPP for Clarification Need Prediction in context-grounded multi-turn Conversation. Clean implementations of QPP baselines suitable for multi-turn conversational dataset with ranked documents (opt.). Designed to detect ambiguous search queries.

Research framework for query-adaptive search: comparing BM25, dense, hybrid, and reranked retrieval under relevance–cost tradeoffs and cross-domain evaluation.

A Python pipeline for ranking and qualifying company profiles against structured and reasoning-heavy business search queries.

Official code for "Session-Guided Pre-training for Query Understanding: Constraining Mask Prediction and Improving Query Fine-tunability" (EMNLP 2026 Industry Track).

Agentic AI search engine with ReAct reasoning and multi-source RAG. Routes queries across Wikipedia, ArXiv & the web using intelligent classification. Built with LangChain & Streamlit, free-tier first.

A reference implementation of modern search architecture that prioritizes deterministic intent enforcement (BM25 + boosts), bounded semantic expansion, and explainable ranking. Built to reflect real production search systems rather than end-to-end black-box ML.

Conversational vehicle recommendation assistant built on a hybrid RAG pipeline — BM25 + FAISS dense retrieval with weighted fusion, cross-encoder reranking, constraint relaxation, and a fully local Qwen3 LLM. Every answer is grounded in retrieved data and explicitly labeled Exact / Relaxed / Fallback.

A robust Retrieval-Augmented Generation (RAG) system for noisy, multi-intent queries using LLM-based query understanding. Implemented in Python with PostgreSQL and OpenSearch for retrieval and storage.