SEARCHing vs ASKing: Case for natural language question answering
What if you could get the exact answer to your online queries on the first try? Instead of searching through a page of results and then a website of information, what if you got the specific answer to your question as soon as you hit enter?
What if you could ask instead of search?
SEARCHing vs ASKing: What’s the Difference?
Since the advent of the internet and search engines, users have grown used to guessing the keywords needed to find answers. If you’re lucky, one of the top results in your first guess will point you in the right direction. Generally, however, you’ll have to go through multiple results, or pages of results, to find the right one. Even when you do find the right website or document, you are often left digging through the material to find your answer. Whether you’re using a search engine or a website’s search option, searching is frustrating, time-consuming, and often inefficient.
Search experience on California DMV website
With asking, search functions use the searcher’s full question to provide context to the query. Then, after using this question to find a shortlist, question answering functions go through the relevant web pages and documents in the shortlist to find the exact answer, including answers from within documents, rather than just presenting the document itself. These answers could be:
● Exact words, phrases, snippets, or generated sentences
● Lists and/or paragraphs extracted from the relevant documents and/or webpages
● Accepted solutions/answers from forums and support tickets
By using the full question, with your wording and thus your intent, the result is exactly what you’re looking for.
SEARCHing vs ASKing: How Does It Work?
Asking also has an intuitive and technical advantage over searching. When using a search engine, for instance, your query goes through several stages:
You guess a combination of keywords.
The search engine removes stop words, such as “a”, “the”, “to”, “from,” etc.; applies synonyms to the words you used; and performs stemming to transform terms like “running” to “run.”
This transformed query is then used to collect results, completely changing your question and losing your original intention.
Its effect is what you see in the above video. Typical website search offers far too many results and most don’t even answer the question. All too often, as experienced by this searcher, potentially useful results redirect users to other pages, sending them deeper into the website without offering anything near the answer they need.
Asking, on the other hand, uses every word you enter to collect results. By using a deep learning technique to convert queries and documents to word vectors in a “vector space,” the ask engine can correlate words and phrases that have similar contextual meanings and usage. The end result then retains the intention and wording of your original search, offering the answer you need.
As you can observe, asking is more efficient and effective, but it also has practical benefits. When experiencing product or service issues, for example, customers can quickly find the answers they need without needing to call, email, or dig through years of forums. Instead, they can use the website’s intuitive ask function to quickly find the answer they need. And if they do reach out to the brand, support agents can find the solution in just seconds, reducing their Service-Level-Agreement and improving the customer experience.
SEARCHing vs ASKing: Where Are We Now?
In 2007, researchers found that 72% of American web searchers reported frustration and impatience over the inefficiencies of search engine results, an experience dubbed “Search Engine Fatigue.” Over the last decade, search engines have done much to improve the searcher’s experience. With new, better algorithms and advances in technology, searching has gotten easier and more accurate, but it still doesn’t offer the answers and experiences users need. In fact, recent research found that 18% of all Google searches lead to a change in the keywords used and 21% of searches include more than one result click per search.
To offer a user experience that is both effective and efficient, search engines and website search functions need to use intuitive, natural language question answering. Until users can get the exact answer they need after asking a single question, online searching will remain a problem instead of a solution.