Lei Feng Wu Enda: recently the net according to the documents will leave Baidu at the end of April. Almost at the same time, Baidu also announced a further deep integration, Baidu core technology will include NLP, KG, IDL, Speech, Big, Data, component technology platform Baidu AI system (AIG), and appointed Wang Haifeng as vice president of Baidu AI technology platform system (AIG) general manager, at the same time for promotion a member of the Estaff, turn to Baidu group president and chief operating officer Lu Qi.
Wang Haifeng is the authoritative scientist at Natural Language Processing field, is the most influential international academic organization in the field of ACL 50 years of history, the only president (President) of the Chinese people, but also as the young ACL Fellow, is the only Chinese from mainland China ACL Fellow. In addition, Dr. Wang Haifeng is a member of the Chinese information society, the editorial board of the Chinese Journal of information, the Chinese Academy of Computer Science (CCF) senior members of the National Natural Science Foundation of the project review panel members. Prior to this, Lei Feng also arranged for Dr.'s lecture on AAAI2017 Baidu vice president Wang Haifeng: Baidu in the field of NLP have done? ".
Dr. Wang Haifeng attended the media activities are not many, but more active on the Quora. According to Dr. Wang Haifeng, Dr. on the Quora of the five questions and answers to organize the cost of text.
1. What do you think of this career change from a scientist to a president of IT?
I'm fascinated by technology and I'm willing to immerse myself in research. I always believe that technology can change the world. Baidu provides me with an ideal platform, where I engage in technical work can quickly and directly benefit the user. That's why I joined Baidu in the first place. In the first few years of Baidu, I led the NLP, voice, image, data mining, knowledge mapping, machine learning, deep learning and other teams. Later, I realized that the great product will connect technology and the majority of users, in turn, will also promote technological progress. Great products, not only the need for advanced technology, but also the need for outstanding design, excellent marketing and efficient management. So I gradually changed my role, from a simple R & D team leader, to become a member of management. I now lead a team of more than 3 thousand people, including technology, product and marketing members, they are very young, energetic, passionate. We have a common goal: to change the daily lives of people with technology and products.
When in charge of a large business team, I need to develop strategies and goals first, and then set up a suitable executive team. For a large team, good rules and culture, began to become an important factor to support and ensure the operation of the business. At the same time, the major breakthroughs in the field of science and technology, the evolution of user needs, as well as the development trend of the whole society, I have maintained great concern.
2, the next 5-10 years, NLP field will be any progress?
Machine Translation, semantic understanding, Q & A and dialogue technology will have a major breakthrough. These technologies will be widely used, and ultimately change the way people communicate with computers, people and all kinds of hardware devices, and between people.
The development of these technologies will benefit from the following four areas: big data, learning mechanisms, knowledge mapping, reasoning and planning.
Big data。 With the prosperity of the Internet, the amount of data and types are growing rapidly. Even the most traditional business areas are beginning to put the data online. Everything on the Internet, everything in the internet. The value of big data will continue to grow in the field of networking.
Learning mechanism. The development of learning mechanisms will continue, which allows us to learn more from the big data.
Knowledge mapping. Through large data and more powerful learning mechanism, we can build a larger knowledge map to the whole world modeling.
Reasoning and planning. Through a large knowledge map, we can make breakthroughs in the field of reasoning and planning. The ability to reason and plan will inject more intelligence into the NLP system.
3. What are the main differences between Chinese and English in the field of NLP?
From the linguistic point of view, Chinese and English are very different. There is no space between words in Chinese written text, and the grammatical relation is expressed in the order of words. These factors add to the ambiguity of Chinese vocabulary, grammar and semantics, because modern language concepts and principles are more suitable for English than chinese.
At present, the mainstream NLP methods are language independent (language-independent). These statistical or neural network algorithms, according to different applications, are further optimized for a specific language.
For example, in May 2015, Baidu released the first large-scale online nerve Machine Translation system. The basic NMT model is language independent, and outputs very good translation results. In order to further improve the performance of translation, we use the specific language features to optimize the translation system.
4, NLP technology used in Baidu products?
In Baidu, we developed a lot of NLP technology, including knowledge mapping, semantic understanding, content labeling, sentiment analysis, generation, summary, Q & A, Machine Translation and dialogue systems, etc.. These technologies have been used in many Baidu products, such as search, feed (News) and intelligent assistants, hundreds of millions of users for service every day. We will all of these technologies are integrated into a platform called NLP Cloud.
NLP Cloud provides more than 20 kinds of NLP modules and programs to serve Baidu products. Our NLP Cloud service is called 1 100 times a day.
To search for example, a typical NLP module, such as segmentation, named entity recognition, grammatical analysis and interpretation are the basic characteristics of. These modules have been continuously optimized and made a breakthrough. Another typical example of the application of NLP technology is question answering system. A high performance question answering system needs to carry on the accurate semantic analysis to the query statement, constructs the broad coverage knowledge map, and carries on the comprehensive analysis to the web search result. When a user enters a query in the search box, the search engine can provide the answer immediately. Many users also use search engines to search for high relevance information to help make decisions. In this case, sentiment analysis (also known as opinion mining) techniques can help to extract a variety of alternative views, and the aggregation of information available to the user.
All in all, NLP technology is an integral part of all natural language related products.
5, in the next 10 years, the search engine will be how to evolve?
Today, when we talk about search engines, the first thought is the search box and search results. What will the future search engine look like? We don't have the exact answer. But we are happy to have a more powerful search engine, so that we can see, hear and feel in different scenes, different products or different interface.Search will be everywhere.
The first point, a more in-depth understanding of the user's intentions, a more in-depth understanding of the content, and the two more precise match, which will make the search engine more powerful. The understanding of the user's intention is not dependent on a single query, but also depends on a broader search context, including query session, time, place, equipment and user personality characteristics. On the other hand, the scope of understanding of the content is also very wide, the need to better understand the semantics of each part of the content, context, view, as well as knowledge extracted from the content. The matching of intent and content will involve all of the factors mentioned above, making it possible to provide the best results for each query in any given context. In addition,. Most of the user queries will be answered directly or executed.
Second, the search interface will be a lot of new changes
Third, the search will go beyond the scope of the existing search engine. Search will be embedded in a variety of products. For example, the search will be one of the basic characteristics of AI hardware products. In the future, the search will be around us. Accordingly, we will also redefine what can be searched. In addition to the existing indexed content, in the future, services, goods, equipment and data can be indexed, searchable.
For a long time, search engine plays an important role in people's daily life. People's needs determine the direction of the evolution of search engines, and technological progress determines how far this evolution will go.