Neural Encoder-Decoder for Long-Tail Search Query Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search systems struggle to generate related search queries for long-tail queries and slight variations, as they rely on frequency counting approaches that require sufficient query reformulation occurrences, leading to inadequate results for unpopular queries and queries with minor variations.
Innovation Solution
A neural network-based approach for generating related search queries using a recurrent neural network and encoder-decoder architecture that processes query intent vectors to produce related queries in parallel with search results, capturing long-range dependencies and handling long-tail queries effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If frequency counting approach is used to determine related queries, then simple implementation is achieved, but inadequate results are obtained for long-tail queries and queries with slight variations
Solution Approach 1:
The patent replaces the mechanical frequency counting approach with a neural network-based semantic analysis system. The neural encoder-decoder model processes query intent vectors to generate related queries, substituting the simple counting mechanism with an intelligent system that understands query semantics, thereby resolving the contradiction between implementation simplicity and result accuracy.
Solution Approach 2:
The patent changes the fundamental parameter for determining query relatedness from frequency count to semantic similarity. By using neural networks to encode queries into vector representations and measuring semantic proximity in the vector space, the system achieves accurate related query generation for long-tail and varied queries while maintaining computational efficiency through parallel processing.
2Reliability
If neural network approach is used to generate related queries, then accurate results for long-tail queries are achieved, but query processing latency increases
Solution Approach 1:
The patent implements preliminary action by pre-training the neural encoder-decoder model on query logs to learn query reformulation patterns. This pre-training phase allows the model to capture semantic relationships and reformulation behaviors in advance, so that during actual query processing, the model can generate related queries rapidly using the learned representations without requiring complex real-time computations.
Solution Approach 2:
The patent uses copying by creating query intent vectors that replicate the essential semantic information of original queries. These vector representations serve as compressed copies that capture the meaning and intent of queries, allowing the neural model to process and generate related queries efficiently based on these compact representations rather than processing full query texts repeatedly.
3Quantity of substance
If frequency counting approach is used, then sufficient query reformulation occurrences are required, but this excludes long-tail queries and new queries
Solution Approach 1:
The patent replaces the frequency-based mechanical system with a semantic-based neural system. The encoder-decoder model learns to generate related queries based on semantic understanding rather than occurrence frequency, enabling it to handle long-tail queries, new queries, and queries with slight variations that have insufficient reformulation occurrences in the training data.
Solution Approach 2:
The patent changes the critical parameter from query frequency to semantic similarity. By encoding queries into vector space and measuring semantic proximity, the system can identify related queries regardless of their frequency in the logs. This parameter change enables the system to generalize to long-tail and new queries by finding semantic relationships rather than relying on statistical frequency.
Data Source
AI summary
A neural related query generation approach in a search system uses a neural encoder that reads through a source query to build a query intent vector. The approach then processes the query intent vector through a neural decoder to emit a related query. By doing so, the approach gathers information from the entire source query before generating the related query. As a result, the neural encoder-decoder approach captures long-range dependencies in the source query such as, for example, structural ordering of query keywords. The approach can be used to generate related queries for long-tail source queries, including long-tail source queries never before or not recently submitted to the search system.