A Large Google Image Search Bug

Here's a large Google Image Search bug. These searches contain two keywords: 1) an instance, and 2) the exclusion of a broad including category.

Search for roses excluding flowers:
rose -flower
16 out of 20 of the images displayed are roses.

Search for ants excluding insects:
ant -insect
20 out of 20 of the images displayed are ants.

Search for circles excluding shapes:
circle -shape
18 out of 20 of the images displayed are circles.

The bug is the following: Google Image search does not understand simple semantic relationships.

The image search algorithm does not know a rose is a flower, an ant is an insect, and a circle is a shape.

Some Large Google Image Search bugs

Here are some large Google Image Search bugs.
Bug #1: Page Rank for images

In Google Image Search, run this search: ice cream. The first results page displays 9 out of 20 images containing ice cream cones.

In my opinion, other types of ice cream probably should have equal or higher page rank:

• ice cream cakes
• ice cream bars
• ice cream sundaes
• ice cream sandwiches
• ice cream shakes
• ice cream scoops

See my earlier post: Is Google Page Rank supposed to be objective?

Bug #2: Incorrect image exclusion operator

Exclude vanilla from your ice cream search results by entering these keywords: ice cream -vanilla.
Here are the results:

The first row incorrectly contains mainly vanilla. The exclusion operator should exclude, not include, vanilla.

A similar bug appears if you try to exclude chocolate.
Try this search: ice cream -chocolate. Here are the results which is a bug:


The results incorrectly display chocolate.

Try this search to exclude cones: ice cream -cone. Here are the results for the first two rows of images..

The bug is that 7 out of 8 of the highest ranking images contain cones. Cones are supposed to be excluded.

The image search exclusion operator bug not only fails to exclude, but it may tend to have the opposite effect by emphasizing what it is designed to exclude - at least for the images with the highest page rank.

SSL Encryption Is Now Available For Google Search Keywords

Previously, I wrote a post describing the security and privacy issues that may arise when using clear text Google queries: Cyber oppression and the problem of clear text search engine keyword requests.
A Google R&D post has now announced that Google searches can be encrypted using SSL: Extending SSL to Google search. Google's new SSL-enabled Beta site is here: https://www.google.com.

My suggestion: Currently, when a user clicks a cached snapshot link in Google's search engine results list, a URL displays that contains in clear text both the 1) website URL, and the 2) user's search keywords.

Example:

Search keywords: [ski equipment]

An example clear text URL of an arbitrary website clicked from Google's new SSL-enabled search results: http://www.skis.com

An example clear text URL of the related cached web site clicked from the same SSL-enabled Google's search engine results: http://webcache.googleusercontent.com/search?q=cache:t3SmcuNSLKIJ:www.skis.com/+ski+equipment&cd=2&hl=en&ct=clnk

Google's cached website versions often are not up to date, and they usually do not display an entire web site. However, some users may prefer viewing them to reduce the chance of governments, ISPs, and other potential eavesdropping entities from tracking their website visits.
It would enhance the security and privacy of these users if Google encrypted the URLs it displays for its cached web sites.

Here is my SETI research: http://setilinguistics.blogspot.com

A Small Google Web Search Options Message Bug

Here's a small Google search options bug:

Run any Google search - say, for volcano. Click Related Searches -> Nearby -> Custom Location. An empty edit field displays and under it the following message appears:

"Sorry, we couldn't understand . Please try another location."

Here are reasons this is a bug:

1. The message should display after, not before, the user enters a value, clicks Search, and Google doesn't understand it.

2. The anthropomorphic "Sorry, we" may be considered a small bug: 1) Google doesn't have to express regrets, 2) why is Google referred to as a plural "we"?

Would Microsoft Windows display a message such as: "Excuse us, but we could not open your dialog box"?

3. There's an incorrect space between the word "understand" and the period.

My suggestion is to improve the message by changing it to something like this:

"Good afternoon. This problem can only be attributable to human error. Please try another location."

Here is my SETI research: http://setilinguistics.blogspot.com

A Small Google Advanced News Bug

The Google Advanced News web page Author field provides this example: e.g. John Smith or "John Smith". The trivial bug is that the two suggested searches are not equivalent.

The first query runs this search author:John author:Smith and finds news articles that have John Smith, John, or Smith as authors. The second query runs this different search author:"John Smith" and finds news articles that have John Smith as an author.

Small Google Calculator bugs

Here are some small Google Calculator bugs.

Bug #1:

The Google Calculator results of these arbitrary calculations include a space instead of nothing or a comma:

32*128
32 * 128 = 4 096

33*27*8

33 * 27 * 8 = 7 128

44+7^4
44 + (7^4) = 2 445

The reason it's a trivial bug is that a user may copy and paste one of these results into another equation and get an error because Google Calculator inserted an erroneous space.

Bug #2:

If you perform a Google Calculator operation, under the computational result Google displays a link asking you whether you want to search for documents containing the terms you entered in the search box.

Example:
(823^2)*5/9082
((823^2) * 5) / 9 082 = 372.896388

Search for documents containing the terms (823^2)*5/9082.

You might expect the suggested search to return documents containing the calculation:

(823^2)*5/9082
However, if you click the link Google Search runs this erroneous query:

823 2 * "5 9082"

Google threw away the "^" and "/" operators, probably changed the "*" multiplication operator to a string wildcard operator, and for unknown, spurious, reasons it changed 5/9082 to an exact phrase by surrounding it with quotes.

Google Rich Snippets: Google Page Rank algorithms vs. user votes

A Google post describes Rich Snippets: "Brief annotations you see beneath search results that summarize what's on a web page."

If a Rich Snippet, say, for a recipe contains a rating property with a high value (e.g. a 5-star review) to what extent does it influence Google's page rank? At its inception, Google could have created a purely democratic "voting" format for web page relevance - however, it didn't. Instead, Google opted for using complex page rank algorithms to determine page rank.

A purely democratic search engine would work as follows: The more people who vote for a web page, the higher it is listed in search query results. All other page rank algorithms can be ignored. 

Improved Google Suggest Interface

A Google post, This week in search 3/26/10, describes a new Google Suggest feature. Now "Google Search" and "I'm Feeling Lucky" buttons display when the Google Suggest box is open.

My suggestion: Make the "I'm Feeling Lucky" button accessible from the keyboard. As far as I can tell, a user can't access it from the keyboard.

Search Engines Promoting Free Expression

A Google post describes a decision to stop censoring Chinese search requests. See: A new approach to China: an update.

Nations handle free expression and censorship differently within their legal structures. It shouldn't be a surprise that search engine requests fall into the constraints of the local laws.

I would like to see a website with a matrix of countries and brief descriptions of how they treat free expression and censorship. It would be nice if it also had, by country, descriptions and links to censored websites.

Here is my original SETI research.

Cyber Oppression And The Problem Of Clear Text Search Engine Keyword Requests

A Google post discusses the crucial problem of threats to online free expression. See: Securing online freedom.

In my opinion, as long as search engine keywords are sent in clear text, governments, ISPs, and other entities can easily match user IP addresses to search requests. Clear text search engine requests can be easily used to identify and eliminate political opposition and also ripoff intellectual property.

Here is an example of a typical Google clear text search query: http://www.google.com/search?hl=en&q=firstopenkeyword+secondopenkeyword+thirdopenkeyword&btnG=Search&aq=f&aqi=&aql=&oq=

Suggestion: Google, Bing, and Yahoo should use SSL to encrypt search queries to help protect their users from political harassment and intellectual property theft.

Is Google Page Rank Supposed To Be Objective?

Web page rankings are determined by algorithms that include a large number of objective and subjective components.

I consider a search engine to be similar to a librarian or a restaurant or a movie reviewer offering a combination of objective and subjective suggestions.

Recently, there have been some Google-related antitrust issues in the news. In my opinion, the more Google claims page rank is an objective, scientific standard the easier it may be to prove its search results are noncompetitive.

Can an organization such as the US National Institute of Standards and Technology measure and standardize web page relevance? 

In my opinion, they probably could not because page relevance includes objective and subjective attributes.

Suggestion: Google should adopt a position that its page rank is similar to a 1) Michelin Guide, 2) Miss USA, or 3) Oscar Awards committee decision.

Here is my original SETI research.

Google Moderator, Citizen Participation, And The Wisdom Of The Crowds

Google Moderator enables organizations to receive and select questions from audiences of any size. It uses a "wisdom of the crowds" paradigm of democracy. Questions for a virtual town hall style meeting are selected based on the number of votes they receive.

However, citizen participation is illusory when thousands of questions are submitted and only a few are answered. In addition, democracy supports minority and individual rights to be heard.

In live town hall meetings, questions are not voted for instead they are usually selected randomly ("You in the second row, do you have a question?"). Follow-up questions are almost always allowed.

Google Toolbar Translation Bug

The Google toolbar enables a nice translation feature. You can hover your mouse over some text and it will translate it.

See: Google translate.

Here's a small Google translation bug: If you try to translate a Google Suggestions list entry that is partially formatted in bold, the resulting translation is only for each separate segment of the word.

For example, if you enter in Google search "resta" one Google Suggestion entry is restaurant. The translation tool attempts to translate "resta" and "urant" separately. However, they aren't actually words.

Google Advanced News Search bug

The Google Advanced News search page has several bugs.

Go to the Google Advanced News search page. In the with at least one of the words field enter this:

"Airedale terrier" "Tabby cat"

These keywords are designed to run a search for articles about Airedale terriers or Tabby cats. Click search. On the results page, the search box incorrectly displays this:

"Airedale OR terrier" OR "Tabby OR cat"

Google inserted the literal OR into two exact phrase match keywords. The search engine now needs to find matches for either of these highly unlikely exact literal phrases: "Airedale OR terrier" or "Tabby OR cat".

I've never seen an animal called an "Airedale OR terrier" or a "Tabby OR cat".

On the same results page the "results" line displays a different incorrect query:

Airedale | terrier | Tabby | cat

This search would look for results with any of these kewords: Airedale or terrier or Tabby or cat.

Click the Advanced News search link to go back to the Advanced News search page.

In the with at least one of the words field a third inconsistent query is displayed:

Airedale | terrier Tabby | cat

This query has now deleted the OR between the terms Airedale | terrier and Tabby | cat and displays "|" symbols that probably should not appear in the with at least one of the words field.

In the Advanced News search page, click search. Now the search box on the results page displays this bug:

Airedale OR | OR terrier OR Tabby OR | OR cat

On the same page, the results line displays another different and incorrect query:

Airedale | OR terrier | Tabby | OR cat

The Google Advanced News has added many bugs to the original query in the with at least one of the words field which was this:

"Airedale terrier" "Tabby cat"

Note that these bugs are different from the bugs I mentioned here:

• Google Advanced Web Search bug
• Google Advanced Image Search bug

Here is my original SETI research.

Google Advanced Web Search bug

The Google Advanced Web search page has several bugs.

Let's say you are searching for information about green peppers or red tomatoes or yellow bananas. In the ordinary Google Web search enter this query:

"green pepper" OR "red tomato" OR "yellow banana"

Click search. The results display Web sites referring to green peppers, red tomatoes, or yellow bananas.

The bug appears if on the results page you click the Advanced Search link.

The Advanced Search query builder comes up with this bug:

| red-tomato | yellow-banana "green pepper"

The leftmost space and " |" symbol do not have any meaning. The query now requires either a "red tomato" or a "yellow banana" and always requires a "green pepper".

If you click search, display the results, and again click the Advanced Search link, the Advanced Search page comes up again with another incorrect query:

| yellow-banana green-pepper "red tomato"

All of the terms have now been made exact required phrases.

Repeat the process and you get this bug:

green-pepper red-tomato "yellow banana"

The Advanced Search page has quite seriously departed from the original query:

"green pepper" OR "red tomato" OR "yellow banana"

Google Advanced Image Search bug

The Google Advanced Image Search page appears to have a bug in its query builder.

Let's say you want to display images of apples or cherries.

Go to the Advanced Image Search page and enter the keyword fruit in the Related to all of the words field. This keyword will disambiguate fruit images from Apple computer images.

In the Related to any of the words field, enter two keywords: apple cherry. Click search. The results display mainly images of apples or cherries based on the search query: fruit apple OR cherry.

The bug appears if you decide to modify the results to include, say, images of avocados. You now want to display images of apples or cherries or avocados.

If you click the Advanced Search link to return to the Advanced Image Search page, add avocado in the Related to any of the words field, then click search, Google runs this query:

fruit apple | cherry apple OR cherry OR avocado

(Note that Google uses both the "|" and the OR symbol to indicate OR).

Since apple | cherry erroneously remains in the query, all results must contain either an apple or a cherry. The last term (apple OR cherry OR avocado) may not return many images of avocados because the results already must include an apple or a cherry.

It is possible, that after the search algorithm runs the term apple | cherry it completely ignores the next term apple OR cherry OR avocado because it has already satisfied the query by returning an apple or a cherry.

You may suggest a user should delete the apple | cherry term in the Related to all of the words field. However, it was written into the field as a query builder by the software based on the user entering keywords apple cherry into the Related to any of the words field.

Note that Google Advanced Web search handles this same situation correctly.

Here is my original SETI research.

Searching Google Blogs

The Google Blog Directory is where you can find out about Google's new products and features.

"Whether it's a product or feature launch or a cool new initiative, chances are that you'll read about most news from Google on one of our blogs."

On the Google Blog Directory web page, Google blogs are categorized by tabs such as Google-wide, Products, Ads, and Developer. You can select a tab and display links to a set of Google company blogs such as the Public Policy Blog, Online Security Blog, and Research Blog.

Suggestion for improving the Google Blog Directory page: There is no search box on the page - how about adding a search box?

(Currently, you need to search in a specific Google blog, and then some of the blogs display a search box with options: This Blog Google Blogs Web Blog News.)

Finding The Laws That Govern Us

A Google post, Finding the laws that govern us, describes a new feature that enables users to read U.S. federal and state court opinions on line.

See: Google Scholar.

You now can use the feature to search by the name of a well-known case, such as Roe v. Wade or Miranda, or you can search by an interesting topic such as the "right to vote", or "freedom of the press".

My suggestion: Digitize legislation and court opinions from many countries in the world and also enable language translation.

This would allow users world-wide to search for topics such as "right to vote", or "freedom of religion" and see and learn how the myriad legal structures of the world compare.

Here is my original SETI research.

Rethinking How To Select Optimal Keywords For Internet Ad Auctions

Most internet ad buyers probably apply a commonly held belief that search volume equates to consumer interest so they bid on the most popular keywords that represent their product line.

A Google.org post appears to refute this strategy, at least in regard to swine flu search volume trends.

See: Google Flu Trends expands to 16 additional countries.

The Google post states that "An important aspect of Google Flu Trends is that we filter out terms that may be popular because people hear about them in the news.

What we do not use in the models is a term like (swine flu) since people are more likely to type that into Google because they want to know more information about it, given the news headlines, and not because they actually have H1N1 or swine flu."

If popular keywords such as "swine flu" do not correspond closely to real world activities (e.g. CDC ILI contacts), it may be preferable for internet ad bidders to invest in R&D to find other sets of keywords that have a stronger correlation with consumer activities and bid on those instead.

For example, a cola soft drink manufacturer may find that its yearly sales pattern is a closer match to Google Trends graphs for "sunscreen" or "ice cubes", than to "cola".

Users may run searches for "cola" to get more information about it, making it a high-volume keyword, but not because they are going to consume it.

Rethinking the keyword search volume to consumer action correlation is a significant paradigm shift. It is already applied in Google Flu Trends, but probably not yet by most bidders in Google ad auctions.
 
See: Search engine ads: auctions or page rank?

Here is my original SETI research.

Predicting Search Trends Based On Historical Data

A Google R&D post describes analyzing historical search trends patterns to predict future search trends. See, On the predictability of Search Trends.

Predicting search trends based on historical search trend patterns is a difficult challenge. There may be insurmountable obstacles embedded in these sayings: 1) Expect the unexpected, and 2) The only thing that is constant is change.

These are my previous posts about Google Trends:

http://googcomments.blogspot.com/2009/08/public-sector-todays-hot-trends.html

http://googcomments.blogspot.com/2009/01/google-trends-comparing-ice-cream.html

http://googcomments.blogspot.com/2008/12/comment-on-letters-to-nature-detecting.html

http://googcomments.blogspot.com/2008/11/google-trends-measures-what.html

http://googcomments.blogspot.com/2008/06/does-google-trends-data-equate-to.html