Search Ranking Compensation via Position Bias Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for ranking search results based on item clicks ratios are inaccurate and unfair, as they do not accurately reflect user satisfaction, leading to sequence inaccuracy and unsatisfactory query results due to the use of empirical compensation factors applied universally across all search rankings.
Innovation Solution
Calculating compensation factors for each ranking position by dividing the position clicks ratio of each ranking position by the first position's clicks ratio, and using these factors to recalculate the ranking scores of search items, allowing for more accurate re-ranking of search results tailored to specific query words.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If item clicks ratio is used to rank search resulting items, then the ranking reflects user satisfaction directly, but the ranking becomes inaccurate due to position bias where top-ranked items receive more clicks regardless of quality
Solution Approach 1:
The patent implements a feedback mechanism where the initial ranking position and click behavior are fed back into the ranking system. The system collects click data from users on search results at different positions, uses this feedback to calculate position compensation factors, and then adjusts future rankings accordingly. This closed-loop feedback resolves the contradiction by using actual user behavior data to correct the position bias inherent in simple click ratio metrics.
Solution Approach 2:
The patent transforms the ranking parameter from a simple item clicks ratio to a corrected ranking score that incorporates position compensation factors. By changing the parameter definition to account for position bias, the system maintains measurement precision while improving reliability. The transformation formula adjusts the raw click ratio by dividing it with the corresponding position compensation factor, thereby removing the distorting effect of position on click probability.
2Measurement precision
If manual adjusting with position compensation factors is applied to all search result rankings, then the sequence inaccuracy problem is reduced, but the compensation becomes inaccurate because a single set of factors cannot suit all query types
Solution Approach 1:
The patent segments the compensation factor calculation by query word, creating distinct compensation factor sets for different queries. Instead of using a universal set of position compensation factors for all search results, the system calculates and stores separate compensation factors for each query word based on its specific click behavior patterns. This segmentation allows the system to adapt to the unique characteristics of different query types, resolving the contradiction between achieving accurate compensation and maintaining adaptability across diverse queries.
Solution Approach 2:
The patent applies local quality by making the compensation factors specific to each query word rather than uniform across all queries. Each query word receives tailored compensation factors that reflect its particular user interaction patterns and position bias characteristics. This localized approach ensures that the compensation is accurately adapted to local query conditions, thereby improving both ranking precision and query-specific adaptability simultaneously.
3Productivity
If empirical compensation factors are used for all ranking positions, then the implementation is simple and fast, but the compensation is too coarse and causes ranking inaccuracy
Solution Approach 1:
The patent applies preliminary action by pre-calculating position compensation factors for each query word based on historical click data before actual search ranking occurs. The system analyzes past user behavior patterns, computes the compensation factors in advance, and stores them for rapid retrieval during live search operations. This preliminary computation resolves the contradiction by separating the computationally intensive factor calculation from the time-critical ranking process, thereby achieving both high precision and fast execution speed.
Data Source
AI summary
The present application provides method of searching information. The method comprises: performing a searching request based on a searching request including one or more query words from a user terminal and initially ranking all the searched items; evaluating the position click ratio of each ranking position with respect to the query words; calculating a compensation factor of respective ranking position by dividing the evaluated position clicks ratio of each ranking position by the evaluated position clicks ratio of the first ranking positions; obtaining a ranking score of each initially ranked item by dividing the item clicks ratio of each initially ranked item by the compensation factor corresponding to the ranking positions of the initially ranked item; and re-ranking all the initially ranked items in descending order based on the ranking scores and then returning the re-ranked items to the user terminal.