How Search Engines Rank Images Using User Ratings

A system that improves image search results by combining how well an image matches a search term with how highly users have rated that image elsewhere.

What it covers

This system improves search results by calculating a final score for images based on two distinct factors. First, it measures relevance by comparing search keywords to metadata associated with the image. Second, it calculates a quality score based on user ratings from various image forums, which is determined before any search query occurs. The system normalizes these ratings across different forums to ensure a five-star rating on one site is comparable to a ten-point scale on another. Finally, it combines these two scores to re-rank the search results, ensuring high-quality images appear higher than low-quality ones even if their relevance to the search term is identical.

What it doesn't cover

• —Does not cover ranking systems that rely solely on keyword matching without incorporating external user ratings.
• —Does not cover real-time ranking adjustments based on user clicks or behavioral data collected during the search session.
• —Does not cover ranking methods that fail to normalize ratings across different platforms.
• —Does not cover systems that lack a pre-calculated quality score independent of the specific search query.

The clever bit

The system treats the 'quality' of an image as a static, pre-calculated property that is independent of the search query, allowing the engine to mathematically merge objective quality with subjective relevance.

Why it matters

This patent addresses the fundamental problem of search engine 'spam' and low-quality results. By shifting the focus from simple text matching to incorporating community-driven quality metrics, it helped search engines provide more useful results. It reflects the transition of the web from a static collection of pages to a social, community-rated ecosystem.

Real-world examples

• 1.Google Image Search results
• 2.Bing Image Search
• 3.Pinterest search ranking
• 4.Stock photo site search algorithms