Deng Yongqiang Zheng Fang
Huge market demand
Through natural language communicate with computer, obtain appropriate information, and satisfactorily, is the goal of people's long-term pursuit. The development of natural language understanding, allowing text automatic classification, intelligent search, information extraction, language translation, automatic abstract, automatic exploration, smart chat, etc.
Natural language understanding also does not meet the same level of intelligent understanding like people, but as long as it delineates specific areas (such as weather, sports, simple information query, or making friends, shopping, health, etc., is wider intelligent chat and consultation, etc.) Energy is tailoring to make a sufficiently practical intelligent information service system, automatically understand the relevant domain issues issued by customers, making information queries more convenient, fast and accurate.
In 2003, international Microsoft, AOL, Yahoo, Google and other companies have begun to provide personalized intelligent information services such as smart chat robots and smart search. Domestic Tencent, HC and other information providers also grasp the international trend in time, launch similar new business, and strive to expand larger market share.
After the information service market has burst, the smart SMS service is in the market in France and China. At the end of 2003, Orange France adopts DEJIMA's technology, providing French SMS users with intelligent information inquiry services such as catering to automotive maintenance. In April 2004, Beijing proudly encompassed the Chinese natural language understanding technology and SMS services, and cooperated with the "Boutique Shopping Guide" newspaper with rich living information resources to provide Beijing catering smart SMS inquiry, and will be more abundant And practical intelligent information services.
Intelligent information services will bring a new application model to effectively improve the quality and satisfaction of services and expand the development space of information services. It will become an important application of the next generation of the Internet and may bring changes in the information service industry. The results of agency show that in 2004, China's SMS market will reach 30 billion yuan, and network SMS users will reach 26.4 million; to 2007, SMS, online chat market scale will reach 150 billion yuan; and market share of intelligent information services It is expected that will reach 10%, which is more than 15 billion yuan.
Rich intelligent service
Intelligent information services have been successfully applied to customer service, knowledge management, intelligent software help, enterprise management, online marketing and other fields. With the high-speed development of the Internet, intelligent information services continue to utterance and popularize, smart SMS services, smart chat robots and smart search engines, etc. have become new application highlights. These service modes are closer, more attractive, and more market influence and value creativity.
Smart SMS service
Smart SMS service is to allow SMS users to enter text directly on mobile phones (instead of entering complex code or cumbersome menus), the system can quickly and accurately understand the user's meaning, provide thoughtful Information and service. SMS service providers can customize a series of new intelligent SMS services, such as tourism transportation, food and entertainment, financial securities, dating network, intelligence quiz, to stimulate users' interest and frequency using SMS services. For example, the Beijing catering smart SMS service provided by the company can be flexible and convenient to guide the user to find the right catering information (Figure 1).
User SMS: I want to eat roast duck.
System Answer: Hello, eat roast duck
There are a lot, I can't send it to you, set a place.
User SMS: Is there near the Xizhimen?
The system replied: there are only 1 in the Xizhimen, Tonghuayuan Hotel Restaurant (No. 101 Northern Lishi Road). If you are not satisfied, you can tell me what you want to eat.
User SMS: I heard that Quanjude is very good.
The system replied: Quanjude roasted duck shop is 32 Yantou Street. Huawei Quanjude is on the 20th of Matsuya. Quanjude and Pingmen Shop No. 11, South Pool Street, Dongcheng District. Thank you for your support, I will find me in the future! Intelligent chat robot
Currently, there are no mature natural language understanding techniques, and only basic, simple dialog exchanges and information queries are achieved. The intelligent chat robot introduced by natural language understanding can understand the needs of users through the form of user dialogue and chat; it can also use intelligent search technology to collect, analyze the information content of the Internet and the knowledge base, perform automatic filtering, screening, Get effective content, intelligently edit the relevant information content, and return to the user. Smart chat, smart game, personalized news customization, intelligent e-commerce, etc. will bring great convenience and more fun to users.
System: Hello, welcome to use this flight booking system!
User: I want to go to Shanghai from Beijing.
System: Which day you want?
User: Is there anything tomorrow afternoon?
System: Are you going to Hongqiao Airport or Pudong Airport?
Users: I want Hongqiao, Air China 757 or 747.
System: Check 3: 13: 30, 15: 40, 16: 20, where do you want?
One class?
User: The latest.
System: Do you want to book a ticket to Shanghai Capital Airport to Shanghai Hongqiao Airport?
User: Yes.
System: Your booking number is 123456, the fare is 850 yuan. Please pay the ticket at the airport or the Civil Aviation ticket office before takeoff. I wish you a pleasant journey!
The above is a dialogue process that specializes in providing a smart chat robot for flight information query services.
Intelligent search engine
The current normal search engine is only matched by keywords because of the keyword techniques, and a large number of information returned is not required by the user. Although the portal or professional website is well classified, it is first to remember the URL, but also to classify the standard, then click the corresponding classification link layer by layer to obtain the information you need. Such information classification query has brought great workload, service costs and entry thresholds, which are also very inconvenient for users. The intelligent search engine is based on semantic network, Chinese word, syntax analysis, handling synonyms, etc. natural language understanding, to a greater understanding of the user's information, get more easiest, more accurate, and smarter. It has the characteristics of humanized, interactive, identifies and answers the user's problem, get rid of the traditional search engine based on keywords, and can provide relevant, other content with reference value.
In addition to the conventional terminal applied to the Internet, the smart search engine can also support the WAP protocol and applied to the phone. In fact, users often need to query information at any time in the outdoor mobile environment of shopping, travel, etc., and mobile phones are very good query tools in these situations, but their screen is small, less memory, low computing performance, narrow bandwidth, not suitable for receiving and saving A large amount of information is more difficult to look at a lot of information satisfactorily. Traditional keywords - based search or classified information query models have certain defects, very unlike applications on mobile terminals. The smart search engine can provide mobile phone users with information services that are free, anywhere.
A typical application of the intelligent search engine is based on rich daily information knowledge bases, build intelligent living information service systems, support Internet, SMS, WAP and other communication methods, in film and television entertainment, consumer shopping, outdoor sports, catering, healthcare, etc. The field provides information exchange means for merchants and users. For example, when the traditional search engine queries the person name, it may have a rename of hundreds of different identities; and the smart search engine can support industry classification functions. When entering "Singer Dawn", you can find directly related to Hong Kong Singer Dawn. Website, as well as his song download address, singer information, news, lyrics, etc., less than two searches, will not point to the novel "The dawn quiet quiet" Support technology
Computers and artificial intelligence technology are the foundation of intelligent information services, and natural language understanding is a support platform for intelligent information services.
The earliest research field of natural language is the machine is machine translation. The technical level of the translation system developed earlier is low and does not meet the practical application. By the early 1970s, the research on language understanding dialogue system has made progress. After entering the 1980s, machine learning research was active and there were many higher levels of practical systems.
Two major technical directions
At present, the natural language understanding technology has two major research directions, which are based on rules, namely, "rationalism"; and analytical methods for large-scale spectrals, so-called "empiric". The former basically mastered the analysis techniques of a single sentence, but it is difficult to cover a comprehensive language, especially for the understanding of the entire paragraph or chapter. The latter makes full use of the computer's high-speed processing capabilities and mass storage, collecting a large number of related text established corpus. The knowledge provided by the corpus is expressed in probability, and thus often answers, and the corpus needs a lot of time and resources to establish, maintenance, and cost. In the future, these two technical directions will take each other to make a long additional short, showing the development trend of fusion.
In order to achieve various functions of intelligent information services, people have developed language data resources such as electronic dictionary, corpus, and other language data resources, such as electronic dictionaries, corpus, and other language data resources, such as electronic dictionary, corpus, and other language data resources.
Pursuit of excellent systems
The natural language understanding system should have good applicability, and it should support the characteristics of context-related analysis, topic free transformation, human machine mixing lead, and speaking dialogue.
Context-related analysis refers to the system that can be understood by the system in understanding the current statement, so that even if there is a certain omitting, the system can understand even if the user said.
Topic Free Transform refers to allowing users to continue to conversion between multiple topics, the system can also remember the previous conversation content. For example, when the user is asking flight take-off time, suddenly insert the destination and local time difference, and then continue with the issue of the flight, the system can answer. Some of the current systems are often only limited to a topic. Once the user changes the topic, the system will be at a loss.
Human-machine mixed dominant is a complete natural dialogue, and users can return to the original topic after converting topics (like people); users can "answer more or less answer); and system can be based on actual The situation extracts semantic information. If the information in the user inquire is rich enough, the system answers the problem directly; if the information is not all or the user is too late, then the system actively inquires to get enough information. The system that does not have a man-machine mixed dominant performance can only wait for the user to ask, if the user doesn't know what to ask, the system will always standby.
The oral conversation is the difficulty of understanding the technology of natural language understanding, but it is also a key point of application system applicability. In the oral language, people's language is very casual, can be omitted, corrected, reversed, and more, these oral phenomena is difficult to solve the traditional single-based understanding of the understanding system, and the introduction of key semantics can be solved well.
From the current statement and technology status, general, high-quality natural language processing systems is still a longer-term effort. But the system for certain specific applications has appeared. Natural language is very complicated, and humans don't know much about the mechanism of yourself. To make your computer understand the natural language, you must first study how people learn and understand the mechanisms of language. In this regard, computer technology and linguistics are required to study cross disciplines.
Research status
In 1963, in 1963, an early natural language understanding system has been built, many of the famous universities and research institutions have invested a large number of scientific and technological resources for research and development. IBM, Microsoft and other companies have launched various products and applications based on natural language-based understanding, and have accumulated a large number of experience and success in English, French. In the 1990s, IBM and Microsoft have established a research institute in China to carry out research on Chinese language understanding, some of which is expected to be introduced in the market in the near future.
China built two Chinese natural language understanding models in 1980, all of which were implemented in a man-machine dialogue. At present, the Units of Tsinghua University, Peking University, Beijing Language University, Shanghai Jiaotong University, Fudan University, China University of Science and Technology, Chinese Academy, Social Sciences, etc. are engaged in the technical research on natural language understanding. Overall, there is a certain gap between natural language understanding at home and abroad. Of course, domestic research also has advantages, and the humanity of experts and language data resources needed by language projects are rich, and the price is relatively low. Some systems have achieved considerable economic benefits. For example, Huaying Group's intelligent auxiliary translation system is an excellent application of language understanding technology; Chinese intelligent input method and other Chinese intelligent input methods introduce a Chinese language model, currently occupying a certain market share; Chinaren (now Sohu M & A), HC and other search engine technology developers have also launched a Chinese intelligent search engine; privileving the company's recently opened smart SMS service system, opening up a new field of Chinese language understanding technology.
Prospect
Human-machine interaction is a very intelligent information service: anyone, at any time, in any place, on any device, you can easily transfer information through natural language and speech, and communicate anytime, anytime. ).
In the near future, we may often have such application scenarios: a person to a strange city (such as Beijing), if you want to learn about Beijing's living service information as soon as possible, you can use your mobile phone or call to provide intelligent information services. system. He can use the spoken to ask a server with massive information: What are the fun places in Beijing? Which places can provide good catering services? Which places can you shop? He can also perform human-machine dialogue with the system, easy to buy tickets, book a room, and so on.
The Olympic Games held in Beijing in 2008 will be the stage of intelligent information service exhibition. Olympics faced language barriers, the industry at home and abroad for this purpose are working closely together to develop "for the Olympic Games multi-language intelligent information network system", and strive for the number of athletes from around the world to millions of reporters, the audience provides integrated, comprehensive, Multi-language, customizable intelligent information services. The system will include multilingual information synchronous distribution, information query, and voice interactive e-commerce, and focus on location-based information services; there are other spoken translation machines and automatic translation phones.
The number of people in Chinese and the distribution range are second only to English. Therefore, the development space of Chinese language understanding and intelligent information service is very large, and the domestic community also recognizes the importance of Chinese information processing, and is carrying out the technology with complete independent intellectual property. R & D, and actively lead the relevant international and domestic technical standards to strive for the initiative of industrial development.
