Key N-gram Extraction for Search Relevance Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing relevance ranking models face challenges in accurately calculating the relevance of less popular web pages with insufficient information, known as tail pages, which are difficult to rank effectively due to limited anchor texts and search-query log data compared to popular head pages.
Innovation Solution
The method involves extracting search-focused key n-grams from electronic documents, using learning to rank techniques to train a key n-gram and phrase extraction model based on search-query log data, and employing these features in a relevance ranking model to improve the ranking of both head and tail pages by leveraging statistical properties and relative importance of n-grams without requiring query and document segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional relevance ranking models use manually tuned parameters or basic learning to rank techniques, then the ranking performance for popular head pages is maintained, but the ranking accuracy for less popular tail pages with insufficient information deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-training the key n-gram extraction model using search-query log data from head pages before applying it to tail pages. This pre-training establishes statistical properties and relative importance of n-grams that can be transferred to tail pages, enabling accurate relevance ranking even when tail pages have limited anchor texts and search-query log data.
Solution Approach 2:
The patent introduces key n-grams as an intermediary element that bridges the information gap between head and tail pages. By extracting and utilizing key n-grams from search-query log data, the system creates a mediating representation that captures the relative importance of different n-grams, which then serves as a foundation for ranking tail pages that otherwise lack sufficient information.
2Productivity
If the model focuses on extracting key n-grams from head pages with rich information, then the ranking performance for head pages is improved, but the ability to rank tail pages effectively deteriorates
Solution Approach 1:
The patent applies universality by creating a key n-gram extraction model that serves multiple functions: it improves ranking for head pages while simultaneously enabling effective ranking for tail pages. The model's ability to identify key n-grams based on statistical properties from search-query log data makes it universally applicable across different page types, regardless of their popularity or information richness.
Solution Approach 2:
The patent utilizes parameter changes by adjusting the model's focus from raw information quantity to the relative importance of n-grams. By changing the parameters used for evaluation (from absolute information content to statistical significance of n-gram patterns), the system can effectively rank both head pages with rich information and tail pages with limited information.
3Measurement precision
If the relevance ranking model relies on anchor texts and search-query log data, then popular head pages with rich information achieve high ranking accuracy, but tail pages with limited data suffer from low ranking accuracy
Solution Approach 1:
The patent applies the taking out principle by extracting key n-grams from the available search-query log data and anchor texts, separating the essential information needed for ranking from the redundant or less important data. This extraction process allows the system to concentrate on the most relevant n-grams, improving ranking accuracy even when the total quantity of available data is limited, as is the case with tail pages.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An n-gram and/or phrase extraction model may be trained based at least in part on search-focused information mined from a search-query log. The n-gram and/or phrase extraction model may extract key n-grams and/or phrases from retrieved electronic documents based at least in part on features and/or characteristics of the key n-grams and/or phrases and based at least in part on features and/or characteristics of the search-focused information. The extracted key n-grams and/or phrases may be weighted. A relevancy ranking model may be trained based at least in part on the information extracted by the n-gram and/or phrase extraction model. The relevancy ranking model may provide a relevancy ranking score for electronic documents listed in a search result based at least in part on weights of extracted key n-grams and/or phrases.