A Study of the Techniques of Automatic Abstracting and Knowledge Acquisition Systems

ing; automatic knowledge acquisition; machine learning; natural language processing Abstract One of the most important signs of the information society is the explosion of information. The information in Internet is out of order and is mostly written in natural languages which need to be processed by the technology of natural language processing. When you search for some certain information on Internet through a search engine, you might be confused by the huge amount of results which the search engine provides. However, if a search engine is embedded with Automatic Abstracting (AA) processing systems, you could locate the information quickly or you could get more information within a limited time. So, the AA technology is valuable both in science and application. The work of this thesis was begun when we took over a project that is called 'The Key Technology Research of Computer Networks Providing Intelligent Information Services' which belongs to the national 863 plan. One of the tasks is 'The Key Technology Research of Automatic Abstracting Systems of Chinese Text'. As a member of this research group, I took part in designing and implementing an AA system called Literature Abstract and Digest Information Extract System(LADIES). From then on, I have been working in this field and this paper is the conclusion of my work. The main topic of the thesis is AA technology. There are two parts of it. One is about the research of understanding based AA systems, and the other is about the invcestigation of Automatic Knowledge Acquistion(AKA) in AA systems. In the first part, the contents of AA technology are introduced and an understanding based AA model is put forward. Based on this model, LADIES is implemented. There are two major features of LADIES: (1) it understands text with the grammar, semantic and pragmatic information of words; (2) it chunks words into a relatively independent entity with chunking rules which are substitutes of syntactic analyzing rules. The results demonstrate that it performs better than those statistical based AA systems. However, the application of LADIES is limited for its knowledge bases. And it is difficult to use in other fields because the knowledge bases are setup manually. So we investigate the techniques of automatic knowledge acquisition in order to solve the above problems to some extent. In the second part, we introduce the basic ideas of AKA and some Machine Learning (ML) methods which AKA applies. Then we propose a comprehensive dictionary model that contains grammar, semantic and pragmatic information of words. And we investigate a strategy of automatic learning pragmatic information for words. Also we put forward another strategy of automatic learning rule of salience sentences in texts and based on it, we establish an AA system LADIES NEW. Eventually, we suggest a AKA based AA system model called hierarchical feature extracting AA system model.

The research and realization about automatic abstracting based on text clustering and natural language understanding

A method of realization of automatic abstracting based on text clustering and natural language understanding is explored, aimed at overcoming shortages of some current methods. The method makes use of text clustering and can realize automatic abstracting of multi-documents. The algo- rithm of twice word segmentation based on the title and first sentences in paragraphs is investigated. Its precision and recall is above 95 %. For a specific domain on plastics, an automatic abstracting system named TCAAS is implemented. The precision and recall of multi-document’s automatic ab- stracting is above 75 %. Also, the experiments prove that it is feasible to use the method to develop a domain automatic abstracting system, which is valuable for further in-depth study.

A Study of Applications of Multiagent System Specifications and the Key Techniques in Automatic Abstracts System

Abstract In this thesis, multiagent system specifications, multiagent system architectures, agent communication languages and agent communication protocols, automatic abstracting based on multiagent technologies are studied. Some concerned problems of designs and realization of automatic abstracting systems based on multiagent technologies are studied, too. Chapter 1 shows the significance and objectives of the thesis, its main contents are summarized, and innovations of the thesis are showed. Some basic concepts of agents and multiagent systems are studied in Chapter 2. The definitions of agents and multiagent systems are given, and the theory, technologies and applications of multiagent systems are summarized. Furthermore, some important studying trends of multiagent systems are set forward. Multiagent system specifications are studied in Chapter 3. MAS/KIB—a multiagent system specification is built using mental states such as K (Know), B (Belief), and I (Intention), its grammar and semanteme are discussed, axioms and inference rules are given, and some properties are researched. We also compare MAS/KIB with other existing specifications. MAS/KIB has the following characteristics: (1) each agent has its own world outlook; (2) no global data in the system; (3) processes of state changes are used as indexes to systems; (4) it has the characteristics of not only time series logic but also dynamic logic; and (5) interactive actions are included. The architectures of multiagent systems are studied in Chapter 4. First, we review some typical architecture of multiagent systems, agent network architecture, agent federated architecture, agent blackboard architecture, and Foundation of Intelligent Physical Agent(FIPA) architecture. For the first time, we set forward and study the layering and partitioning models of the architectures of multiagent systems, organizing architecture models, and interoperability architecture model of multiagent systems. Chapter 5 studies agent communication languages. We study FIPA Agent Communication Language(ACL) message grammar, Communicative Act(CA) types, ACL semantics and Agent Communication Protocols. The string grammar of ACL messages and SL grammar represented by EBNF are given, and a new ACL semantic model is set forward. We study automatic abstracting based on multiagent technologies in Chapter 6. For the first time, the idea of constructing automatic abstracting systems based on multiagent technologies is set forward, and the necessity and feasibility of constructing such systems are set forth. The architecture of automatic abstracting systems based on multiagent technologies is provided, and the basic elements and mutual relations are studied. We set forth the working principle of the automatic abstracting systems based on multiagent technologies and how the contradiction of domain currency and abstract quality of automatic abstracting systems is resolved. Studies, simulation analysis and design of the automatic abstracting experimental system based on multiagent technologies(ABES/MAT) are made in Chapter 7. We study the architecture of ABES/MAT, and set forth its basic elements and mutual relations. The user model and mathematics descriptions of ABES/MAT in the Internet environment are provided. Three coordination algorithms that synthetically use environment information, system information and history information are set forward. Then, we analyze the performance of ABES/MAT based on the simulations, and get suitable number fo abstracting agents for each domain in a given load. Finally, we compare the three coordination algorithms. The realization of ABES/MAT is introduced in Chapter 8. Firstly, we study the problem of the agentification of legacy systems, and show how to realize the agentification of glanc, ladies and news. Then, the architectures and functions of Coordination Management Agent(CMA) and Information Agent(IA) are set forth. The agent naming grammar represented by EBNF is given., In the following, we explain the designed agent communicative acts and management