A Time and Season for Search: How Data Mining Can Influence Search Advertising

With more than 3 billion search queries a month, a search engine like Yahoo might be tempted to take a close look at, and analyze the data it receives in its search logs. That data might tell it what people tend to search for at different times of the day, and different days of the year. The search engine may also be able to tell sometimes whether those searches were performed by men or women, by people in different locations, and may look at other information they might have about those searchers.

That analysis, that collection of data, might be helpful in deciding what to show searchers in advertisements, and in other content displayed to people looking for information.

A flow chart showing different kinds of user data that could be analyzed to identify popular products and concepts that could be used to target advertising and the display of content on the Web.

A patent granted to Yahoo last week describes how the search engine might collect data over time, looking back possibly more than 10 years, to decide what to show people who view its advertisements, and possibly other content upon its pages, and the pages of publishers who display its ads. The patent is:

Temporal targeting of advertisements
Invented by Anand Madhavan and Shyam Kapur
Assigned to Yahoo
US Patent 7,672,937
Granted March 2, 2010
Filed: April 11, 2007

Abstract

A system and method utilize temporal targeting of content, such as advertisements. The targeting may be based on time of day, day of year, season or upcoming holidays.

In addition, prior search history may be utilized to determine current popularity and/or predict future popularity for a particular concept that may be used for targeting.

Yahoo shows sponsored ads along with search results when someone performs a search at the search engine, and it also offers advertising on publishers’s websites through its Publisher Network. The ads shown tend to be based upon the queries used to perform a search, or upon the content of a publisher’s page. But what if there isn’t an advertisment available that matches up with those queries or content?

Might information about what people may interested at different times of the day and different days of the year be helpful in providing ads on those pages?

If Yahoo were to spend time analyzing the query terms that people search with to find patterns within those searches, it might tell them that people tend to look for products or articles related to sleep problems at night, or information related to the stock market in the morning. The search engine might identify when people begin performing queries about costumes and candies in the days before Halloween.

If Yahoo were to analyze queries in real time, or near real time, it might be able to notice when certain query terms started to spike in popularity in a short durations of hours. The search engine might also look back through the years, even up to 10 years ago according to the patent, to make note of trends related to holidays and seasons.

Targeting of ads or other content based upon time, date, or season, as well as an analysis of prior browsing or searching history over a certain period of time could suggest topics and items that are likely to be the most popular and/or relevant for targeting.

Historical User Data Used to Determine the Temporal Popularity of Topics

The patent tells us that the search engine might look at a wide range of information collected in a search log database relating to user behavior to determine what concepts or topics might be useful in deciding which advertising and other topics to show. The description of the patent tells us that the search engine might use the process described for more than targeting advertising, though the claims listed in the patent seem to focus primarily upon targeting ads.

Here are some of the types of user data that might be analyzed for this targeting method:

Searches – Search queries may illustrate patterns over time to suggest topics that may become popular.

Page Browsing – The pages and items on a page that a browser may view or scroll over.

Product clicks – The ads that someone clicks through and product pages someone views.

Purchase history – Actual purchase histories of goods and services.

User inputs – The kinds of information that someone might enter into a website.

User profiles – Information entered by someone to create an online profile, such as sex, birthdate, location, and/or interests.

We are told that other sources of information gathered about online activities might also be considered as well:

Any additional information may be used as an input type for determining the popularity of topics or content. For example, news, blogs, and/or entertainment may provide data for predicting the popularity of certain concepts.

For example, if historical data suggested that whenever a hurricane was discussed in the news, then users viewed content on insurance, it would suggest that insurance is a popular concept whenever a hurricane is in the news.

Likewise, blogs or entertainment, such as movies, television shows, or sports may provide additional data for determining concepts that may be popular. News about entertainment and the blogs may be monitored to determine which topics are the most talked about.

Conclusion

I started this post noting that Yahoo receives more than 3 billion queries a month, and I took that number from the home page of Tapas Kanungo, who notes on his page that the snippet generation process he worked on while at Yahoo was responsible for showing snippets for approximately that number of queries a month.

Analyzing that many searches can provide a great amount of information about what people are interested in at any point in time. It’s common sense to assume that people are more likely to be interested in snow shovels in the late fall and winter, and swim suits and sun screen in late spring and summer.

But, having a large amount of data that can unveil patterns involving other topics and interests that might span hours or a decade could provide some insightful views of what might interest searchers. Using that information to target advertising might make it more likely that people will click on advertising when there aren’t ads to show that are appropriate for the content of a page, or the queries used in a search.

I suspect that information could be used in other ways as well, though the patent discussed in this post doesn’t explore those other possibilities in much detail.

Share

15 thoughts on “A Time and Season for Search: How Data Mining Can Influence Search Advertising”

  1. 3 Billion Queries a month for Yahoo? This I did not know thanks Bill. Honestly so many people talk about Google, and for good reason. I have completely ignored most of the other search engines lately and just monitored my Google analytics. I know that just in looking into numbers of searches, keywords etc, can be overwhelming at times. Have you found a good program for analyzing these things in an efficient way and presenting them concisely? Another great post thank you Bill!

  2. I agree with your conclusion that “Analyzing that many searches can provide a great amount of information about what people are interested in at any point in time.” This will really be a big help to advertisers. They would now have that great timing in promoting products, especially with regards to seasonal products.

  3. Hi Andrew,

    We get most of our information about how much traffic or business a search engine might get through things like Comscore estimates from a small percentage of self-selected internet users who agree to have their traffic monitored in exchange for things such as free anti-virus software or the chance to win a sweepstakes. So, when we see numbers straight from someone working at one of the search engines about how busy they are, it’s worth noting.

    Analytics can be helpful in deciding upon actions to take, and possible changes to incorporate on a web site, but they can also be intimidating if you don’t have an idea of what to do with the numbers that you are seeing. I’m not sure that any tool can replace by itself the ability to apply thoughtful independent analysis on the data you see before you. Rather than a specific program, I’d definitely recommend spending some time over at Avinash Kaushik’s blog, Occam’s Razor.

  4. Hi Andrew,

    I think that the kind of analysis described in the patent can help the search engine’s decide when to show certain kinds of advertising or content, but I don’t know how they would feel about presenting that data itself directly to advertisers. As a potential advertiser, I would love to see that kind of information directly, but I don’t know how feasible or likely that would be.

  5. I have to agree with Andrew on this.

    The first thought that popped in my head was how this could directly benefit me as an advertiser. It makes complete sense how analyzing the mountains and mountains of data available to Yahoo would enhance the usability for both the searcher and advertiser.

    Thanks Bill, I’m loving your stuff here!

    Jason

  6. Hi Jason,

    The hard part of using all that data is knowing which of the data is useful, and which isn’t. I know that the search engines collect an incredible amount of information about visits to web pages, queries that people use, sites that toolbar users browse, and more. Picking the right data to use could be very helpful. Knowing which data is the right data is the real challenge.

  7. Hi Clinton,

    It’s hard to tell if Yahoo might consider sharing this kind of information, but we do seem to be seeing more of that kind of sharing from the search engines, such as Google showing us Google Trends and Google Insights for Search. Interesting thing about those is that they aren’t selling the information, but rather providing it for free.

  8. I am looking forward to the new search engine Blekko. They will supposedly have all the data us webmasters are looking for when it comes to search queries. I don’t like using Yahoo because of the ads they have in the results, to much for me. Great post! :)

  9. Hi Lisa,

    I’m a little behind in my reading on Blekko, but I did manage to catch a little of the news. It does seem like they want to be more transparent about how they rank results, and share more about how their system works. I’m wondering if that is a good idea or not. I guess we will see.

  10. But how many searches does Google has, if Yahoo is much smaller than Google it will be enormous. For the advertisers it is great that the search engines know what you like. Also Google gives personalised adds, I’m looking for a new phone. Every time I see a google adsense bar I see the type of phone I’n looking for.

  11. Hi Arvid,

    I’ve been seeing some Adsense ads that seem to be more related to what I’ve been searching for and viewing then they are to the content of the pages that they appear upon as well.

  12. Hi Bill hope you are doing well. It has been a while since I have been to your site for a conversation. So how do you feel about Yahoo and Bing becoming one? It doesn’t seem like they will cut into Google’s market share all that much.

  13. Hi Bill,

    Good to hear from you.

    I think it’s still early to say how Yahoo and Bing joining forces will impact the search field.

    I do wish that Yahoo had found a way to thrive as a search engine on its own. They do have some interesting patents and some fairly talented engineers working with them.

Comments are closed.