x
Systems question - answer (in English " question answering systems ," or systems QA ) are regarded as the next step in the evolution of search engines the information . QA-type systems are characterized by the fact that they receive a set of questions in natural language and, based on a collection of documents, must extract the answer or answers. This collection can range from a simple local collection to the entire World Wide Web .
Systems for extracting answers to questions in natural human language fall into the category of information acquisition systems.
We can characterize as knowledge the state of an information system coupled with an environment. By information system can be understood both an individual (in the sense of a person seen as a distinct unit from other people) who interacts directly and consciously with the environment and who builds through their own means (observation, cognition, etc.) the universe of knowledge , but also a technical system that receives information from various human agents and processes it or makes it available to other individuals. The Internet can be described as such an information system that, interacting with the environment through human agents, builds its own knowledge base.
Knowledge accessible through the Internet is, in most cases, already in the form in which human agents make it available to the system, namely in natural human language.
Today, the Internet is, without a doubt, the largest knowledge base, in a continuous expansion and refreshment. It is, at the same time, one of the most accessible places where this knowledge can be consulted. But the degree of development of the Internet also has negative aspects: due to the multitude of information available, finding the necessary information at a given time can be difficult or even uncertain.
The most effective methods of discovering and acquiring information are currently search engines. Their purpose is to provide the user with a set of articles or web pages in which he can find the information he needs. Often the articles offered by search engines do not fulfill the user's desire to get a satisfactory answer. Also, they do not provide the concrete answer to the user's problem, but only a set of web pages, from which the user must extract the information sought.
The next step in the field of information acquisition is the development of systems capable of answering questions posed by the user in natural language. The main aim of such a system is to provide an answer to the user's question that meets the following three conditions: to be correct, to be formulated in natural human language and to be succinct enough. A question-and-answer system requires more complex natural language processing than document acquisition systems.
In theory, natural language processing is a very attractive topic, due to its applicability in areas such as human-machine interaction. In practice, however, there are a number of major difficulties, mainly due to the different way in which a statement in natural language can be interpreted and the multitude of meanings that the constituent words can take. Question-answering systems, seen as a subdomain of natural language processing, inherit their problems.
The first question-and-answer systems were developed in the 1960s and were, in fact, just interfaces that used human language to query expert systems created for different fields. Among the systems of this type we mention:
The mentioned systems were very efficient in the fields in which they operated. LUNAR was presented at a conference in 1971 and was able to answer 90% of the questions asked by people who were not specially trained to work with this system. A main feature of these systems was that they all had the information stored in a database, which had to be developed by experts in that field.
SHRDLU and ELIZA followed . SHRDLU was a system capable of answering questions about its own universe, constructed of various geometric shapes, which the system could move, and ELIZA was able to simulate a conversation with a psychiatrist. Both systems were closed systems in terms of information conveyed, based, in practice, on a closed universe.
In the 1970s and 1980s a number of systems were developed that could interact with the user using natural language. The most important are Unix Consultant (a system that answered questions about the UNIX operating system) and LILOG (it was able to provide tourist information about a German city).
The heyday of question-and-answer systems, however, took place in the late 1990s, when the Text Retrieval Conference (TREC) included a section dedicated to such systems. TREC is a conference dedicated to research in the field of information acquisition, initiated by the National Institute of Standards and Technology (NIST) and Advanced Research and Development Activity (ARDA) - a department of the US Department of Defense , which takes place annually and began in 1992. .
Modern answering systems, developed under the impetus given by TREC, are broad-based systems. This requires that the database considered must be much larger than the databases for the first systems, developed in the years 1960-1970-1980. A large number of articles (for example, articles in encyclopaedias, newspaper articles, blogs, etc.) are considered as a source of information, accessible in particular via the Internet.
An interesting achievement of a question answering system is the START project , developed at MIT by InfoLab Group. Like most systems of this kind, START also uses English as a working language. Another implementation is AskJeevs .
Also, important companies that develop information search solutions on the Internet have projects regarding the development of such systems (eg: Microsoft , Google , IBM , etc.).
There are two options for building a question answer system:
The choice of one of the two models depends on the complexity of the questions to be asked and the degree of performance desired by the system. It is clear that systems in the second category are superior to the first.
If the beginning of artificial intelligence in the 1960s, scientists were fascinated by the idea of being able to build systems capable of responding to questions belonging to limited areas (closed domains), currently development of the Internet and the steps taken in the recovery information (information retrieval - IR) and techniques of natural language processing (natural language processing - NLP), and the demand for easy access to information, has increased interest in systems that provide answers in broad areas (open domains).
A question-and-answer system based on a collection of documents typically has three main components:
1. The question analysis module - transforms the questions formulated in natural human language into queries for the document acquisition engine;
2. Item Purchasing Module - searches the item collection for items relevant to the user's question, based on data received from the question analysis module;
3. The answer extraction module - from the collection of items returned by the article purchase module, extracts a concise answer that is the answer in natural human language to the user's question. If such an answer does not exist in the document collection considered by the item acquisition module, it is preferable that the system does not answer the question, instead of returning an incorrect answer.
Even though the Internet is a medium full of information in all areas, finding an answer to a simple question can sometimes be a difficult task. Some of the difficulties that may arise in the development of such a system:
