Neural Query Auto-Correction Using Beam Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current search engine query completion methods are inadequate in handling errors and unseen permutations, as they rely on memory-based schemes that cannot generalize suggestions beyond stored data and fail to provide real-time, accurate completions.
Innovation Solution
A neural network-based system that combines character-level language modeling with an edit-distance-based potential function and a tree-based beam search algorithm to generate and rank query completion suggestions, enabling error correction and real-time processing without the need for extensive storage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If memory-based completion schemes are used, then real-time performance is achieved, but the system cannot handle error correction or unseen query permutations
Solution Approach 1:
The patent segments the query completion task into two distinct modules: a neural network-based error correction module that handles spelling errors and generates corrected queries, and a separate completion module that generates completion suggestions. This segmentation allows each module to specialize in its function while working together to achieve both real-time performance and error correction capability.
Solution Approach 2:
The patent introduces an intermediary error correction layer between the user's input and the completion generation process. This intermediary module corrects spelling errors in the input query before passing it to the completion system, enabling the system to handle erroneous inputs while maintaining real-time performance through efficient neural network processing.
2Speed
If database lookup methods are used, then fast completion is achieved, but the system requires extensive memory storage and cannot generalize to unseen queries
Solution Approach 1:
The patent replaces the traditional mechanical database lookup system with a neural network-based completion system. Instead of storing and searching through large databases of query permutations, the neural network learns patterns from training data and generates completions probabilistically, dramatically reducing memory requirements while maintaining fast completion speed.
Solution Approach 2:
The patent changes the fundamental parameter of how completions are generated: from deterministic database lookup to probabilistic neural network prediction. This parameter change allows the system to generalize to unseen queries by learning underlying patterns rather than relying on pre-stored permutations, reducing the need for extensive memory storage.
3Reliability
If neural network-based completion is used, then error correction and generalization are achieved, but real-time performance and low latency are compromised
Solution Approach 1:
The patent performs preliminary action by pre-training the neural network on large datasets of queries and their corrections before deployment. This pre-training enables the network to quickly process and correct errors in real-time during actual use, as the complex learning work has already been completed during the offline training phase.
Solution Approach 2:
The patent implements a dynamic two-stage processing system where the error correction module and completion module operate sequentially but efficiently. The system dynamically adjusts processing based on input needs, correcting errors first then generating completions, optimizing the balance between accuracy and latency through streamlined neural network operations.
Data Source
AI summary
Various embodiments of systems and methods allow unsupervised, deep learning, character-level language models to complete and correct search engine queries, given arbitrary search prefixes. Embodiments use a multi-layer, character-level, recurrent neural network trained on query logs to generate query suggestions for the user. Integrated is an error correction framework. More particularly, approaches disclosed herein for query error correction and completion combine the character-level language models with an edit-distance-based potential function calibrated to generate spelling corrections, linking the two using a tree-based beam search algorithm that can exploit the potential functions and efficiently rank the suggestions. Optimizations are made to the predictive system, and efficient processor-based computations complete the queries, with error correction, in real-time. The embodiments substantially increase hit rate over standard approaches and are capable of handling tail queries.