Start-up brings Magi search engine

Absrtact: How would you react if a start-up team told you that their startup project was to do a search engine? Last week I received a project called Magi, the team behind the search engine, the mammoth Explorer and Rasguea

How would you react if a start-up team told you that their startup project was to do a search engine?

Last week I received a project called Magi, the team behind the search engine, the Ji Shu of mammoth browsers and Rasgueado and the peak Labs he founded. So what's so special about Magi compared to ordinary search engines?

To put it simply, Magi is a new natural language search engine + knowledge map service. Ordinary search engines, whether Google or Baidu, can only return to the user a number of possible links to relevant web pages. To solve the problem of user search, users also need to click on their own web links to dig and find really useful information. And Magi helps users skip this link, giving answers directly to questions based on natural language search, rather than web links. This answer is accompanied by a visual logic map, and there will be some reference links (when the product is getting better, consider no page links at all). Ji Shu says Magi is similar to a civilian version of IBM Watson or Wolfram Alpha.

Having said such a big paragraph, I decided to use one of the most "intuitive" examples to tell you Magi's special place. If you want to know the truth: how many breasts are there in Cang Jing? What are the actresses with bigger breasts? You can put this problem directly in the Magi search box, and then a carriage return, Magi will tell you directly, is G, and then listed a chest than she also a big name list of actresses.

This example is mentioned in the following video (by the way, the actress is larger than the original number of actresses).

' The example that makes the answer sexier is really hard to achieve, Ji Shu said. Magi is the use of machine learning to solve problems independently. In this process, magi through Natural language processing + web crawler, understand the user's problem, and then learn from the massive data and return to the user. To complete these tasks in a short time, the algorithm is very demanding. To ensure that the answer is reliable, magi each input issue to the standard knowledge map, probability statistics, neural network of these three different strategies, and then to compare and vote, and finally determine an answer.

What's the difference between Magi and IBM Watson and Wolfram Alpha?

Ji Shu First said the common point of the three:

Are serious knowledge logic engine services. They're not amusement-quiz bots, for unanswered questions, neither Alpha nor Magi will reply to your nose-picking, or personal assistant, who will not help you set your alarm or sing a joke; it is not a web App portal or a box calculation, and there will be no non-unitary results rendered, There is also little human-directed optimization intervention (at least Magi, Alpha has a certain labor cost), and all inputs generate output that can serve as an intermediate value using a unified expression structure and automated logical processing of local data.

For the difference, Ji Shu explains this: Wolfram Alpha is the computational engine (computational Engine) + a certain general knowledge, the essence of the knowledge of computing, Watson is stronger than information processing, and Magi, the essence of data learning. Another important difference is that Magi is for Chinese. Chinese natural language processing is much more difficult, Wolfram Alpha founder Stephen Wolfram once came to 36 Krypton demonstrated their Chinese features, but the effect does not seem ideal, probably because the great God himself does not understand Chinese.

However, there is a good product is not enough, how to make this more intelligent search engine is also a difficult problem to promote. Ji Shu's idea is to go through two ways: first, they developed a plug-in, as long as the browser installed this plug-in, using Google or Baidu and other search results will automatically embed Magi more intelligent search answers, so that more people know Magi; second, by partnering with Third-party apps, By Magi Free to develop some of the technology to promote their own search services. For example, Magi and a healthy application of the cooperation, is to use open voice search technology to exchange each other's use of Magi search services.

Chris Dixon once wrote a similar view on his blog: As more and more people have terminal devices connecting to Internet portals, a big entrepreneurial opportunity in this era lies in how to share and preserve the spirit of the user to create results (such as wisdom, experience, and knowledge). The form of Wikipedia, question and answer website in fact is only an roughing process, and Magi provides logical structured knowledge, is a process of deep processing.

The more interesting thing about doing this in the form of a search engine is that after adding a user account system, everyone can automatically generate a structured logical knowledge base based on the search history, which is a valuable asset to users. When you search more, the system will help you organize the more perfect knowledge system. At this stage Magi has not done this, Ji Shu told me that their current concerns are mainly user privacy issues.

In addition to the logic of the Knowledge system, Magi actually represents a search engine development trend. Now the internet is getting younger and smaller, and the ability of a new generation of netizens to use search engines through keywords is also slowly deteriorating. So the trend of search should be more human nature language search.

Magi will be in the formal online today, interested students play your wisdom to magi questions!