Conversational Search Reference Resolution System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current search engines poorly handle co-reference resolution, particularly with plural pronouns, partial names, and across domains or platforms, leading to incorrect entity resolution and irrelevant web results.
Innovation Solution
A system that performs co-reference resolution by parsing search queries to identify entities and conversational identifiers, mapping them, and reformulating queries to ensure accurate reference resolution across multiple queries and domains, using a component that tracks previous queries and updates context as needed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If search engines use traditional entity resolution methods, then the system complexity remains low, but the accuracy of resolving pronouns and partial names deteriorates
Solution Approach 1:
The system segments entity resolution into distinct components: pronoun resolution, partial name resolution, and full name resolution. Each component handles specific types of referential expressions independently, improving accuracy without requiring complete system redesign.
Solution Approach 2:
The system performs preliminary entity resolution by maintaining a context stack that stores previously resolved entities. Before processing new queries, the system pre-resolves pronouns and partial names using stored context, enabling accurate resolution without complex real-time analysis.
2Adaptability or versatility
If search engines resolve co-reference across domains and platforms, then the versatility of conversational search improves, but the difficulty of detecting and measuring correct entities increases
Solution Approach 1:
The system implements a universal entity resolution framework that handles multiple types of referential expressions (pronouns, partial names, full names) across different domains and platforms through a single coherent mechanism, enabling cross-domain co-reference resolution.
Solution Approach 2:
The system introduces an intermediary entity resolution component that mediates between user queries and search results. This intermediary resolves pronouns and partial names by matching them against entities in the context stack, simplifying the detection process even across diverse domains.
3Reliability
If search engines maintain context across multiple queries, then the relevance of web results improves, but the loss of time for tracking previous queries increases
Solution Approach 1:
The system performs preliminary context storage by maintaining a context stack that caches previously resolved entities. This preliminary action enables rapid retrieval and matching during subsequent queries, improving relevance while minimizing the time cost of context tracking.
Solution Approach 2:
The system selectively discards expired or irrelevant context entries from the context stack while recovering and retaining important entity resolutions. This selective management maintains search result relevance over time while preventing unbounded context accumulation.
Data Source
AI summary
Methods, computer systems, and computer-storage media are provided for reference resolution. The present invention seeks to resolve entities in conversational search. To enable a more natural conversational search interaction, referential expressions such as pronouns are handled in queries. A component keeps track of previous queries and performs reference resolution based on an entity in the previous query and reformulates the query using an identifier for the referenced entity. Reference resolution may be performed for pronouns, plural pronouns, partial names, across domains, and on any device or platform.