Classification and principles of building question-answer search systems

The analysis of the ambiguity of the natural language is based on the development of Question Answering Systems that can process a user-entered question in a natural language and give a meaningful answer. Unlike the classical keyword search task, in which the result is a list of documents, in the question-answer search task, the result is a short and concise answer generated by the system as a result of analysis of various data sources. The review article lists and considers the main approaches and principles for constructing question-answer systems: a meta-search system, an annotated text search system, an expert system, a search system in question and answer collections. One of the first approaches to Question Answering Systems can be called the BASEBALL system of the early 60s of the last century, since it became possible to ask questions to the system in a natural language, but the knowledge base was a regular structured database. Thus, it can be considered its natural input system. All early Question Answering Systems were faced with the problem of the lack of BigData - a large amount of digitized facts and rules. Really working expert systems were obtained only in a limited domain of knowledge. Therefore, for a qualitative discussion of Question Answering Systems, it is proposed to classify them in the following dimensions: types of supported questions, types of supported answers, source of information, technique for outputting a question or answer by source of information, limited domain of knowledge, quality assessment methods, direction - who asks the question: user or system. Direction is a dimension proposed for the classification of Question Answering Systems for the first time in this article. It defines the lead question-answer dialogue in a pair - a man-computer. A direct question-answer system implies that a person asks questions and the machine answers. An inverted system assumes that the computer is leading this dialogue. Today, one of the most developed and well-known direct Question Answering Systems is the question and answer system on the IBM Watson supercomputer. In the last decade, there has been an active development of educational technologies on the Internet (EduTech). Using the accumulated amount of data on successful or dead-end paths by the user on digitized course materials, it is possible to form an adaptive training course for each of them, which allows to maximize the student’s readiness coefficient. In practice, training is faced with a high threshold of entry by the author of the course, the teacher. He needs not only to restructure the structure of his classical courses, breaking up into smaller blocks, but also to add original questionnaires and elements of gamification of instruction to the pauses between blocks. Inverted Question Answering Systems with an open dialog domain of knowledge have a great prospect in solving the problem of generating original questionnaires, conducting a simple dialogue on an adaptive question graph and introducing gamification elements to improve the perception and assimilation of lecture material.