Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Xinyu AI Search: Enhanced Relevance and Comprehensive Results with Rich Answer Presentations

Bo Tang, Junyi Zhu, Chenyang Xi, Yunhang Ge, Jiahao Wu, Yuchen Feng, Yijun Niu, Wenqiang Wei, Yu Yu, Chunyu Li, Zehao Lin, Hao Wu, Ning Liao, Yebin Yang, Jiajia Wang, Zhiyu Li, Feiyu Xiong, Jingrun Chen

TL;DR

Xinyu AI Search tackles the core challenges of generative AI search—relevance, coverage, and user-friendly presentation—by introducing a query-decomposition graph (QDG) that breaks complex queries into sub-queries and enables stepwise retrieval and generation. The system couples multi-source retrieval, query expansion, and context-aware generation with novel rich-answer representations, including built-in citations, a timeline visualization, and textual-visual choreography to improve comprehension and trust. Extensive online deployment and ablation studies show Xinyu outperforms eight existing technologies in relevance, comprehensiveness, and insightfulness, while demonstrating the necessity of each sub-module. This work presents the first comprehensive framework that integrates retrieval, generation, and user-centric presentation for generative AI search, offering practical improvements and a blueprint for future research and deployment.

Abstract

Traditional search engines struggle to synthesize fragmented information for complex queries, while generative AI search engines face challenges in relevance, comprehensiveness, and presentation. To address these limitations, we introduce Xinyu AI Search, a novel system that incorporates a query-decomposition graph to dynamically break down complex queries into sub-queries, enabling stepwise retrieval and generation. Our retrieval pipeline enhances diversity through multi-source aggregation and query expansion, while filtering and re-ranking strategies optimize passage relevance. Additionally, Xinyu AI Search introduces a novel approach for fine-grained, precise built-in citation and innovates in result presentation by integrating timeline visualization and textual-visual choreography. Evaluated on recent real-world queries, Xinyu AI Search outperforms eight existing technologies in human assessments, excelling in relevance, comprehensiveness, and insightfulness. Ablation studies validate the necessity of its key sub-modules. Our work presents the first comprehensive framework for generative AI search engines, bridging retrieval, generation, and user-centric presentation.

Xinyu AI Search: Enhanced Relevance and Comprehensive Results with Rich Answer Presentations

TL;DR

Xinyu AI Search tackles the core challenges of generative AI search—relevance, coverage, and user-friendly presentation—by introducing a query-decomposition graph (QDG) that breaks complex queries into sub-queries and enables stepwise retrieval and generation. The system couples multi-source retrieval, query expansion, and context-aware generation with novel rich-answer representations, including built-in citations, a timeline visualization, and textual-visual choreography to improve comprehension and trust. Extensive online deployment and ablation studies show Xinyu outperforms eight existing technologies in relevance, comprehensiveness, and insightfulness, while demonstrating the necessity of each sub-module. This work presents the first comprehensive framework that integrates retrieval, generation, and user-centric presentation for generative AI search, offering practical improvements and a blueprint for future research and deployment.

Abstract

Traditional search engines struggle to synthesize fragmented information for complex queries, while generative AI search engines face challenges in relevance, comprehensiveness, and presentation. To address these limitations, we introduce Xinyu AI Search, a novel system that incorporates a query-decomposition graph to dynamically break down complex queries into sub-queries, enabling stepwise retrieval and generation. Our retrieval pipeline enhances diversity through multi-source aggregation and query expansion, while filtering and re-ranking strategies optimize passage relevance. Additionally, Xinyu AI Search introduces a novel approach for fine-grained, precise built-in citation and innovates in result presentation by integrating timeline visualization and textual-visual choreography. Evaluated on recent real-world queries, Xinyu AI Search outperforms eight existing technologies in human assessments, excelling in relevance, comprehensiveness, and insightfulness. Ablation studies validate the necessity of its key sub-modules. Our work presents the first comprehensive framework for generative AI search engines, bridging retrieval, generation, and user-centric presentation.

Paper Structure

This paper contains 66 sections, 2 equations, 5 figures, 16 tables.

Table of Contents

  1. Introduction
  2. The Development and Challenges of Generative AI Search Engines
  3. Our Approach to Generative AI Search
  4. Our contributions can be summarized as follows:
  5. Paper Organization
  6. Terminology
  7. Related Work
  8. Retrieval
  9. Contextual Generation
  10. Orchestration
  11. Technical Approach of Xinyu AI Search
  12. Unified Framework of Data Preparation and Model Fine-tuning
  13. Query preprocessing
  14. User Intent Understanding
  15. Query Rewriting
  16. ...and 51 more sections

Figures (5)

  • Figure 1: Common issues in generative AI search answers.
  • Figure 2: Xinyu AI search framework. The upper row illustrates the full response pipeline, while the lower row provides a more detailed depiction of several novel approaches integrated into this framework.
  • Figure 3: Domain distribution of 300 test queries.
  • Figure 4: Ablation study of sub-modules for the rich answer representation.
  • Figure 5: Xinyu's Interface for query disambiguation.