Search quality insights from Bing's core search Research Department

Intermediary transaction http://www.aliyun.com/zixun/aggregation/6858.html ">seo diagnose Taobao guest cloud host technology Hall

Preface: This is an article from the Bing Core search and development manager, this article deals with the series of features of Bing, which is just a common search feature improvement, but read through this article, we will find that the search engine will be a lot of effort to study the mistakes people make during the search, how to correctly understand the user's intentions and use thesaurus to provide more accurate content. So Lou believes that both Bing, Google and Baidu will do the same.

In Bing, we are committed to providing the best search results, while the research Group data mining experts are constantly providing our core spelling and ranking algorithms, the reality is that there are always some legacy flaws in the history, partly because people are too reliant on search mistakes, and in this article my colleague Bill Ramsey (Bing's research and development manager) will introduce the incidence and severity of defects in three ways: URL queries, tracking links, and related searches.

Web site Query Common errors

This is one of the major sources of search flaws, involving what we call URL queries, such as "facebook.com" or "Yahoo/e-mail" queries, at first glance, you might think this is a simple question. After all, search engine (Bing) contains billions of URLs, and finding a matching site is not much of a challenge. But in reality, this type of query is actually quite complex. Because all of us will use countless spellings and variants.

For example, "facebook.com" has thousands of different variants, such as "Facebookc.om", "facbook.com", and "WW.FACEBOO.OMC", and people do not always know the correct URL except for such spelling errors. For example, Southwest Airlines is southwest.com, but some people try to search "swair.com" to reach the company's Web page. At the same time, we usually see URLs such as "Yahoo site/mailbox" when the correct URL is "mail.yahoo.com" arrangement.

Even if we find out your true search intentions, a malicious URL or spammers present another challenge. They hunt for top-level domain name objects such as coolmathgames.com (people are actually coolmath-games.com) URLs.

This is our flaw, we mainly through three areas to solve this kind of error query method:

First, by properly identifying URLs, we can block the problem by identifying the URLs that we avoid, such as including searscardcom.com garbage results.

Second, the simulation of user error testing, through billions of of the original model, we can solve common misspelled URLs.

Third, we will analyze and look for sites like "swair.com" so that users will end up with a scheduled site "Southwest.com".

Another example, suitable for machine learning mode, like "facebooklogin.com" query equivalent to "Facebooklogin.net", which is a very common domain name suffix input error. In addition, like "Sofitel Bath and beyond.com" input into bedbathandbeyond.com. Our model has adapted to these changes and will proactively modify the search results, and the following examples are Facebook users:

Delete RELATED LINKS for redundant traces

One of the key functions of a search engine is to query the components that perform spelling and query extensions, and the spelling corrects hundreds of wrong queries, while the search for occurrences of the phrase (The following query box indicates that we have changed the user's query), we put this alteration as "recourse". For example, if you enter "stories about successful heroes", we will show "successful heroic deeds including quotes", but we may only show "Heroes and Deeds of success" and we will set up all of your intentions.

In the past, we used synonyms as part of our tracking links, but often led to deviations from the theme of search results, resulting in synonym tracking as superfluous, and all of us expanding the definition of "word" to help its users better match.

So this feature we've removed, the added value is small, when Bing changes some synonyms, tracking links may not be able to add more valuable information, so we change the color of the search results to black. We will continue our efforts to provide better queries for user-specified search terms.

Improve related Search

Related search, this everyone is familiar with, in people's initial search, we will be related to the search to prevent the search results to the left, now adjust to the right, such as search "Brad Pitt":

(Note: Bing Chinese version has not changed, Baidu and Google at the bottom.) )

Sometimes we will query the results of the digression. For example, "AMD" will provide users with unexpected search results, by improving our relevant models.

At the same time, we have made other improvements beyond the format clause in the relevant search, namely "KSN Weather Lab" in "KSN Weatherlab", (note: Some experimental projects are in the beta phase) and avoid retrieving adult content in secure searches.

Conclusion：

A bit of search engine is that it will always depend on people, and people always have different flaws, we do is to reduce the defect rate and search rate, hope that people can do less search and do more things.

Author: Dr. William Ramsey--bing, chief development manager of Core search and development

Article Source: Lou Blog This article address: http://lusongsong.com/reed/488.html