Neural Query Reformulation for Null and Low Search Recall
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search engines face challenges with null and low queries, where the user's vocabulary mismatch with item vocabulary leads to insufficient recall and unsatisfactory results, consuming excessive computing resources and wasting time due to repetitive searches.
Innovation Solution
A neural translation model with a sequence-to-sequence architecture and diversity-inducing optimization function generates diverse and multiple query reformulations to enhance recall and reduce repetitive searches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a search engine uses traditional vocabulary matching methods, then the system complexity remains low, but the recall and relevance of search results deteriorate when user vocabulary does not align with item vocabulary
Solution Approach 1:
The patent replaces traditional mechanical vocabulary matching mechanisms with a neural translation model that uses sequence-to-sequence architecture. This neural system automatically learns and adapts vocabulary mappings between user queries and item descriptions, resolving the contradiction by substituting rigid mechanical matching with flexible neural processing that maintains high recall and relevance without requiring complex manual configuration.
Solution Approach 2:
The patent changes the fundamental parameter of vocabulary matching from exact string matching to probabilistic neural translation. By transforming the matching mechanism into a parameter-based approach where the neural model outputs translation probabilities and multiple reformulated queries, the system achieves improved recall and relevance while managing complexity through automated learning rather than manual rule creation.
2Reliability
If the search engine generates multiple diverse query reformulations, then the recall set size improves by 100%, but the computing resource consumption increases
Solution Approach 1:
The patent applies partial action by generating multiple query reformulations selectively based on query difficulty assessment. Rather than always generating all possible reformulations, the system identifies queries that benefit most from reformulation and applies the neural translation model accordingly, achieving improved recall set size while controlling computing resource consumption through targeted application.
Solution Approach 2:
The patent performs preliminary action by pre-training the neural translation model on large datasets of query-item pairs. This offline pre-training phase captures vocabulary mappings and translation patterns, enabling the system to generate multiple diverse reformulations during online search with reduced real-time computing overhead, thus balancing recall improvement with resource efficiency.
3Loss of time
If the search engine implements repetitive search handling, then user time consumption increases, but the vocabulary alignment problem persists
Solution Approach 1:
The patent performs preliminary action by proactively generating multiple diverse query reformulations and searching for matching items before the user needs to retry. The neural translation model anticipates vocabulary mismatches by producing alternative phrasings of the user query, and the system executes searches with these reformulated queries in advance, eliminating the need for repetitive user searches and reducing time loss while maintaining vocabulary alignment through automated translation.
Data Source
AI summary
A search engine leverages a neural translation model to provide diverse and multiple query reformulations. Two decoders are injected and a diversity inducing optimization function is introduced. After a query input is received from a user, a number of items are retrieved from a database in response to the query input. In response to a determination the query input is a null and low query based on a number of responsive items, a plurality of decoders is injected and a diversity inducing optimization function is leveraged to generate a plurality of diverse reformulated queries. A plurality of query results corresponding to the plurality of diverse reformulated queries is provided as output.