"Editor's note" the writer @ Chen Cun, Cambridge University natural language processing (NLP) master of group. In machine learning, semantic analysis engineer.
Not long ago, Google announced a reorganization of the Android business executives, Google Senior Vice President sangdaer·pichayi became the new CEO of Google. This change seems to have caused some dissatisfaction with Google Now the development team, lead to full separation (also said key members of release), Google Now merged into Search groups. Google is Now the Android platform to counter Apple's Siri, Microsoft Cortana's voice Assistant. So what is Google Now? It contains what technology Google black?
What is Google Now Cath Kidston iPhone 6 cases
Although this article talking about the broad set of App Google voice search is Now integrated, but Google Now more precisely defined narrowly, is Google on the mobile phone application Now open, appear on the first screen of the intelligent recommendation cards, at the right time to give you the right information, such as weather, news. Relative to the active input in Google search input box you want to find something, Google will Now "smart" to push the information you need in front of you.
So often Now referred to as Google Google's smart push technology.
Such as author when in a foreign country, can enjoy the convenience of Google Now. For example through Booking.com (similar to c) book the hotel, or kayak.com (like go) ticket confirmation email will be sent to my Gmail mailbox. Google will be "smart" analysis (peep) mail these trips, and a few hours before the flight about to take off your flight details and Terminal and gate information. When you arrive in a new city, and automatically eject you designed navigation routes, and contact phone number and other information. When traveling abroad, Google Now will automatically pop up the local currency, as well as the country. It will also remember your home carefully and position of the company, reminding you to go home on time traffic during off hours, and not smart enough to understand that you have to work late tonight!
Here say that the birth of Google Now
Google Now, is the legendary product of Google 20% Project. Baris Gultekin Google engineers in 2010 when rich sensors on the personal interest in using cell phones to do some interesting things, the idea coincides with then-head of the Google Map of Andrew Kirmse. Two people in their spare time to build a prototype and show time in MUI Marissa Mayer of Google.
Supported by the MUI, the team was growing, and was one of the users commuting time prediction and the integration of Google Calendar events. As it happens is that at the beginning of 2012, Larry Page requires Android and Search teams to build a similar thing. Support of the Android leader Andy Rubin, the two men joined the Android team, get a lot of design resources, and published for the first time in the 2012 Google IO Google Now.
After so many years of development, Google Now has now been integrated into Google search App.
Recently a foreign science and technology Web site also reported rumors that Google Now team, said to be members of full separation, also linked to the recent Google schema change, conspiracy theories of meaning. Some key members leaving anything, but this happened as early as a few months ago, perfectly normal personnel changes.
Google Now itself as an excellent product, from the Android team into the Search group, from a technical point of view, big Search group can get additional technical resources, early completion of the smart Google of this great goal is good, this is obviously good news!
Although the App interface is simple, just a search box and a card inside the technology won't be easy.
Introducing special Google Now recommended cards, we see Google Now with this little app, and contains a lot of Google's black technology: Smart recommends (Google Now card), hot words awaken (OK Google), voice recognition and semantic analysis.
First a little bit about intelligent recommendation
Recommended principles are "Just right information at right time", that is the right time to recommend the appropriate information. How to know what users really want, rather than machines gave that needs to handle. Although the statistical machine learning has been applied on a large scale in industry in recent years, but although the context of the unknown user information in all aspects of life, be advanced machine learning algorithms, smart recommendation cannot be made too, so companies using a combination of rules + machine learning strategies.
It appears from the existing product, Google Now recommend things are quite conservative.
Google can you stay in a place often guess at your home and work location, and traffic conditions in specified time automatically reminds you to go home every day. Cath Kidston iPhone 6 Case
Mentioned earlier, travel message cards, hotel cards, foreign travel exchange rates and country time cards, and so on. Recommendation algorithm how much difficulty is not high, but Google can collect as much as possible to the user's lifestyle information in dribs and drabs and through thoughtful rules, do not bother to recommend to you. In other countries, because the message high penetration, Google with Gmail (by the way, Gmail with Google Now is born that year, and was born of the legendary 20% project), Google now,Google has its geographical and human.
Hot words awaken
Google Now the main interface of the software, on a cell phone call with a "OK Google", you accept user automatic speech recognition voice queries.
Hot words awaken as the key to start voice recognition, is designed to accept user with power consumption as low as possible awakening.
Because of the need to continue to open the microphone, operating on the audio input, the CPU has not been dormant, will seriously affect the already tense cell phone battery life.
Words awaken on a cell phone in order to maintain high standards, focus of the need to take into account, in addition to accuracy (user name is you can immediately respond, user is not called when you don't listen to the right), as well as energy consumption. Hot word wake not need like General voice recognition as recognition kaleidoscope of natural language, it as long as recognition "OK Google" this fixed of Word, so in algorithm aspects also without to used complex of traditional voice recognition, abandoned tall Shang of various HMM, and language model, with simple directly of classification of method in user entered voice flow in the looking for "OK Google" this several Word, simple but easy.
Of course, there is an exception, Moto x just came out a few years ago, is a 24-hour wake full voice, they have 2 low-power chips, specifically waiting for the user to input voice commands.
Speech recognition, needless to say, Google's strengths
A few years ago Microsoft research a dissertation based on deep learning applied to speech recognition, definition of algorithm accuracy beyond the traditional speech recognition based on HMM-GMM. But Google lead its industrialization, and apply to that year's voice search product. DNN on the original audio data directly to speech recognition, and require a lot of training data, neural network to train a large number of parameters, so the hardware requirements are very high, this is Google's strengths.
After years of accumulation, and Google's speech recognition has begun to take shape. Noteworthy is that as a competitor of Google Now Siri,Siri has been using the nuance of speech recognition technology (nuance is the world's largest provider of speech recognition, can be understood as the fly in China), while Apple also reported some time ago to set up his own message of the speech recognition team.
The semantic recognition
Through the Google search box above Now, you can ask some questions look very cool.
Some of daily life information queries, such as New York, flights to San Francisco, such as the next 3 days the weather in Beijing, Tokyo time. This question-and-answer system involving all aspects of natural language processing, it is very difficult.
Although academically, NLP (natural language processing) in a variety of areas, such as parsing, entity recognition, after decades of effort, has made more than 90% of accuracy. But in the actual context, users more diverse colloquial expression, ambiguity, and academics are often used for evaluating system set of language tests varies widely.
And, question-and-answer system is the product of several NLP system into line, such as speech analysis output as an input to parse, parse input output may be recognized as an entity. In such a system, errors will be propagated to the line of the earlier steps, and was magnified.
Therefore small errors on a single subsystem, will eventually be enlarged, understanding of the serious impact on end users ' queries.
General question-and-answer systems are specifically optimized for current field, study what are the most frequently asked questions, and we have relied on the rule (even a regular expression) and strengthening user guide, reducing the ambiguity of natural language.
Google Now voice search also supports Wikipedia issues, such as current United States President who is, how much is the height of the Himalayas, or how large is the population of China.
Google after all is search origin, sat hold whole Internet of information, tens of thousands of of page in the has contains has "Obama is United States President", "China of population is 1.3 billion" such of statement, for Google for, problem will simplified into how match user of problem to tens of thousands of page in the of a a sentence words, and just to out Top dozens of of match page, this in screen small of phone Shang, and even watches Shang on user experience of effect vital.
Academic general practice is to first identify users are asked which type of problem (What? When? Which? Where? Etc.), and then using information retrieval strategy first filtered in thousands of Web pages and user issues most relevant to candidates page, and according to the type of problem, these candidate pages to pick out the most direct statement that answer user questions. Win Jeopardy! (Can be understood as the old quiz so quiz shows) IBM Watson of the human champion system also uses a similar design. This method ensures the Automation methods to improve the coverage of the problem of non-human, rather like a fruit needs to recruit a lot of problem edits to deal with more and more enthusiastic masses to the relentless teasing of Siri.
Google recently, semantic analysis, mapping knowledge devoted a lot of energy, we can Google Now as well as in general search to see the integration of such technologies.
So, Google Now merged into large groups of Search and get more technical support is definitely a big improvement!
If you are a guest, you can fill in the following form, with the direction of Lei feng's network for a small business survey;
If you are a senior, you can scan the QR code below the table to participate in an opening game, Rango, prepared by Huawei P8, development boards and other gifts.
Young, brave enough to battle!
250 votes
Google Glass
Google Google Glass glasses is developed by Google-an augmented reality wearable smart glasses, glasses will set smart phone, GPS, camera in one can navigate through the camera, video call, voice control, notification information and e-mail address, and so on. Through eyes of the user information in real time, users don't have to do it, they can receive, with your voice on the Google Glass
View details of the voting >>
No comments:
Post a Comment