Google is currently making changes to its core search algorithm, which says it can change the ranking of results in one-tenth of the query. It is based on cutting-edge natural language processing (NLP) technology developed by Google researchers and applied to its search products in the past 10 months.
Essentially, Google claims that it improves results by better understanding the interrelationship of words in sentences. Google discussed an example at a press conference yesterday, and its search algorithm was able to parse the meaning of the following phrase: “Can you buy medicine for a pharmacy?”
Pandu Nayak, a Google researcher and vice president of search, believes that the old Google search algorithm treats the sentence as a “bag of words.” So it looked at the important words, medicine and pharmacy, and simply returned the local results. The new algorithm is able to understand the context of the term “for someone” to realize that it is a question about whether you can accept a prescription from another person-and it returns the correct result.
The optimized algorithm is based on BERT, which advocates “negotiating from Transformers two-way encoder.” Each word in the abbreviation is a term in the NLP field, but the point is that instead of putting a sentence, BERT looks like a bag of entire sentences All words in. This can make people realize that the word “for someone” should not be discarded, but is essential to the meaning of the sentence.
The way BERT realized that he should pay attention to these words was basically achieved by self-study in the Titanic game of Mad Libs. Google extracted a set of English sentences, and randomly deleted 15% of the words, and then the task of BERT was to figure out what these words should be. Jeff Dean, Google’s senior researcher and senior vice president of research, believes that over time, this kind of training is very effective for making NLP models “understand” the environment.
Another example cited by Google is “no street parking.” The word “no” is essential for this query, and Google’s algorithm ignored this before implementing BERT in search.
Solemnly declare: The copyright of this article belongs to the original author. The reprinted article is only for the purpose of disseminating more information. If the author’s information is incorrectly marked, please contact us to modify or delete it as soon as possible. Thank you.