Semantic Search is defined as search for information based on the intent of the searcher and contextual meaning of the search terms, instead of depending on the dictionary meaning of the individual words in the search query.
Let me explain Semantic Search with few real world examples.
Let us say you are working on your computer and someone ask you "Do you have windows there", it means whether you have Microsoft Windows operating system on your computer. But if you are approaching a Realtor looking for office space and ask him the same question, it takes a completely different meaning. In that context, the question means whether the office space you are discussing has any windows and ventilation in the room.
Humans can understand the question based on the context and give relevant answer. But how about search engines?
Search engines cannot differentiate the question based on the context because it does not know the context in most cases or does not depend fully on the context even if it is aware of it.
Take another real world example:
You walk in to the kitchen and ask "It is so hot outside and I am thirsty, can I have something?". The answer could be anything like "Here is some cold milk", "Drink this cold soda" and so on. You may not find anything in a kitchen with the label "Drink if you feel hot" but anything with the labels Milk, Apple Juice, Orange Juice and so on would be just fine. Also, the answer would exclude anything with the label "cooking oil", "dish wash solution", "hot coffee" etc. The answer is provided based on the intent of the person and not based on the words in question matching the labels on the items in the kitchen.
A human can understand the intent and context of a question, but search engines and software systems cannot.
Semantic Search in Search Engines
Semantic Search in search engines mean the search engine would provide relevant search results based on the intent and contextual meaning of the search terms.
Search Engines are getting smarter and smarter. Advanced search engines like Google and Bing are capable of giving search results based on the semantic meaning of the terms, even though none of them are perfect in semantic search.
Semantic Search: Factors considered by Search Engines
Semantic search does not just mean contextual search or search based on the intend of the question. It include several other factors as well. A smart search engine would consider several factors to provide the most relevant and useful search queries, including:
If the president election was just finished in the country and someone is searching for 'Who is the new president', the semantic search system should be able to understand the query and give relevant results based on the current trend and news.
Location of search
If a person is searching for 'what is the temperature', the semantic search engine should be able to provide results based on the current location of the search. If the person is searching from California, search results should include the current temperature in California.
Intend of the search
I have explained this case in length above. Semantic search engines should be able to give appropriate search results based on the intent of the search and not based on the specific words used in the search query.
Variations of words in Semantic Search
Semantic search should consider tenses, plural, singular etc and provide relevant search results for all semantic variations of the words. For example, words like dog, dogs, dog's etc.
Synonyms and Semantic Search
A semantic search engine should be able to understand the synonyms and give more or less the same search results on any synonyms of the word users search for. For example, try searching for "biggest mountain" and "highest mountain". You would get pretty much the same results since both of them means the same in this particular query, even though the "biggest" and "highest" could mean different things in different cases.
Generalized and Specialized queries
A Semantic Searching engine should be able to set relation between generalized and specialized queries and provide appropriate and relevant results. For example, consider an article on general health topics and another article specifically on Diabetes. If someone search for health information, both articles could match even though the article on Diabetes does not talk specifically about "health".
This is a sub-set of context matching in semantic search. Semantic search should understand the broad concept of the query and return relevant results. For example, a query on "Traffic problems in New Jersey" could return relevant results including the topics "narrow roads", "non functioning traffic lights", "lack of roadside assistance" etc because in a broad conceptual point of view, all of these lead to traffic problems.
Natural language queries
Not everyone are tech savvy and not many people know what to search to get the relevant search results. Most users simply type in queries in natural language. For example, if some one want to find what is the current time in Arizona, USA, they would search for 'What time is it in Arizona'. Most search engines would simply show results from the websites and articles that talk about Time and Arizona. However, smart search engines that use Semantic Search would actually show you the current time in Arizona, USA. Try it yourself at Google search.
Change of meaning based on the group of words.
By combining different words, the true meaning of search term could change. Consider the following search terms:
New egg health products
New egg health benefits
If you search for both the above terms in Google, you would get completely different meaning. Instead of just picking the results based on the words, Google Search looks at it as a term and then combines with common user search pattern. The first term returns search results primarily on the popular online shopping website NewEgg.com and shows results of health products from that site and similar sites. The second term shows search results for the health benefits of Egg.
So, you can see that the last word in the search term completely change the meaning of the search query. Search engines that make use of Semantic Search should be able to distinguish such differences and give relevant search results.
Semantic Search: The challenges in front of search engines
Search engines has to deal with many challenges when it comes to semantic search. The first and foremost challenge is, understanding what exactly the user means when it comes to search terms on search terms with different meaning in different context.
Consider this search query: "best window paint"
It is hard to understand whether the user mean the Paint software in Microsoft Windows computers or paint for the windows of a home or car. In such cases, different search engines behave differently, providing significantly different search results.
Semantic Search - Compare Bing and Google search results
Both Bing and Google are very powerful search engines and both of them can handle semantic search reasonably well within the limitations of a non-human system.
I did a quick search for the term in both Google and Bing.
See the results produced by Bing and Google for the same search term "best window paint":
Semantic search results - compare Bing and Google search results
Click on the image to see full size image
You can see from the above image that both Bing and Google understood the search term differently. In case of Bing, it showed a mixed result including Paint software in Microsoft Windows and also the painting of doors and windows of home and automobiles. Bing search results were dominated by results from Microsoft Windows and related topics. But Google understood it as 'painting the windows of a home or car' and all except one result in the top 10 results was about those topics. There was 1 search result which was about the Paint.NET software utility.
This small experiment shows how challenging the semantic search could be for search engines. No matter how smart a search engine is, sometimes it cannot figure out what exactly the user means. To make it worse, the same search term could be used by different people in different meaning.
For example, someone who is planning to paint the windows at his home may search for "best window paint" to find out the best paint available in the market . Another person may look for some painting utility for his computer so he could draw some pictures for the school project.
The above sample is just a single case of Semantic Search experiment. You could search on any number of such terms and find different results in Bing and Google.
Bing and Google are not pure semantic search engines. However, both of the search engines incorporate many elements of semantic search in to their search algorithm to provide more relevant and useful search results.
Semantic Search is a big challenge for search engines and none of them are perfect. Most search engines have improved significantly in last few years. Search engines like Bing and Google provide significantly relevant search results incorporating some degree of semantic search. There are many other specialized search engines which offer purely semantic search results, but they lack many other qualities of normal search engines. I will cover them in another post.
|Article by Tony John|
Tony John is a professional blogger from India, who started his first Weblog in 1998 at Tripod.com. Tony switched to blogging as a passion blended business in the year 2000 and currently operates several popular web properties including IndiaStudyChannel.com, Techulator.com, dotnetspider.com and many more.
Follow Tony John or read 804 articles authored by Tony John
|Guest Author: Andrew 24 Oct 2012|
|Thank you for sharing the information. You have explained everything in very simple words and anyone can understand. I read many articles but never figure out what exactly is semantic search.|
Thanks for the great article.