| According to statistics, there are 120, 000 lawyers, 230, 000 public procurators and 210, 000 judges, undergraduates of 12, 000 law schools of every year at present in our country. Their ability of obtaining and using legal knowledge are different. What merits special attention is that reasoning function of people exceed machine, but their memory function, search speed and accuracy are not as good as machine far also far. Have it is fairness that cases are dealt with that can not get kind settlement. Lawyers, public procurators, judges, legislators, jurists already are not able to bear the legal documents of accumulating over a long period and legal case's heavy load of increasing constantly. Developing legal expert system which can help them reason and search powerful legal is very urgent already. Under Dr. Chen's helping and support who is a doctor of computer department of the Hong Kong Polytechnic University, author of this paper investigate further in the field of legal case's category and judging. the thesis is author's work summary and raise with the theory for master's degree.The difference of legal case document and legal categories is expressed in quantity. Artificial neural network (ANN) and genetic algorithms (GA) is used to the legal field, and it is successful to settle the two difficult points of a legal expert system—Classification and judging of case, which is the foundation of finishing the research and development of legal expert system for the future. The legal expert system that we imagined is used to assist the judging of the case. Divide three pieces of module: Category—give a case document and belong to the corresponding laws; Answering question—every laws corresponds to a series of questions which are researched and proposed together by law experts in advance, and searching the answer for every question in the document; Judging—get the result of adjudicating finally (the probability of wining or losing a lawsuit) through reasoning and calculating according to question's answer.In this paper what is discussed in detail is researching and developing of key technology of categorizing and judging, Shown in the following several aspects mainly:1,Developing the category and judging subsystems of the legal caseThe author finished researching and developing of the categorized subsystem independently, and participated in researching and developing of the judging subsystem. By assessed and test, the accuracy rate of the categorized subsystem is more than 75%, and the accuracy rate of judging subsystem is above 68% too; The speed is very fast, generally within 1.5 seconds; What is more, they have fine maintainability, the portability, expanding, etc. it has reached the anticipated goal.2,Had improved and expanded the categorizing algorithm of the levelResearch of categorizing algorithm is important in the categorizing subsystem. Characteristic words weight formula has been improved especially and put forward the unique basic thought of categorizing. In the categorizing subsystem: First, we gained characteristic words weight tables of each category through a mass of law training-cases documents which have already been categorized. Second, calculating each category's weight summation based on the weight tables of characteristic words. Finally, we distribute the law-case to the category which got the largest summation and at the leafage of the category tree.3,Have discussed the participle algorithmTo offer and comment famous words-parsing arithmetic, and point out their pluses and minuses at present. Put forward a new words-parsing arithmetic based on Law-lexicon, which is the core module of the system. Arrange the structure of the Law-lexicon, and raise the speed and accuracy of the automatic words-parsing of system. At the same time, improve the accuracy of categorizing. 4,Design and use the algorithm of GA-BP.Set up one corresponding ANN in the judging subsystem. Because the ANN algorithm's speed of convergence is slowly and it is possibility to run in the local and the GA can overcome this problems validly, so we adopt the GA and carry on the primary election to the weights of network firstly, and then optimize the weights of network with the Error Back Propagation algorithm. Experimental result proves that the result is correct higher rate than result which is received with traditional statistical analysis.To the field of law case analysis, the article has put forward some new thoughts and algorithms, which were applied to the real system and make better result. But, the research work of legal expert system is very difficult. The key technological of the second module has not been totally solved yet, and still need further investigation. Especially, Study of the respects, such as in accuracy of categorizing, semantic analysis, accurate answer etc, still needs to be strengthened.
|
PDF Full Text Request