Skip to main content

Popularity Doesn’t Always Mean Precision

We have seen in class that Google is able to return search results efficiently through using an algorithm similar to PageRank where a sites reputability is dependent on its hubs. Although this is essentially Google’s game winning formula, it is only a system of looking at connections and returning links that it thinks the searcher wants to see. In the article written by Matt McGee, we see that Google’s image search is not as amazing as Google’s normal search. The direct image comparison is not very accurate as Roz from Monsters Inc. ended up returning some wrestling images. However, when searching using pictures of some well known pictures such as landmarks and common images such as a flower, we see that similar images are returned and even a text description is sometimes returned. This is interesting because PageRank searched for similar images on reputable sites and returned the information/description from those sites. However, reputable sites don’t always have the correct image reference. In fact, many forums, picture libraries or websites not in the SCC may have useful information in classifying images. For example, determining what the picture of the sunflower apparently can’t always be done accurately by looking through reputable websites that have an image somewhat similar to the sunflower. Since image search is also part of Google Search, the returned images and search results may still be biased by PageRank and reputation. When Google Image searching for Roz, there are some Roz pictures, then seemingly completely random pictures, then the occasional actual Roz picture such as one’s DeviantArt drawing. The primary problem is that images and good references may often exist in the upstream or downstream branches of the internet where PageRank will not look first. This article shows that reputation doesn’t always correlate with accuracy and that image search may need a different kind of ranking system.



Leave a Reply

Blogging Calendar

October 2017
« Sep   Nov »