It is an absolute necessity that the kind of technology for analysis of word p roblems. Despite many achievements in this field by scholars home and abroad, t here still exist many problems like shortage of formulas for various word problem s and incomplete analysis from the artificial system. Therefore, further study is qu ite essential in how to develop such a technology for the analysis. A semantic m odel for the analysis of word problems has been proposed by the research group, the basic principle of which is to build equations on the numerical relationship i n the word problem. The paper is aimed at building a set of semantic models to pick up as many numerical relationships as possible in the word problem.And th ese models can be generalized into a triple, including key words; relations; map.H ere, the map refers to the list of maps between strings of keywords and expressio ns for numerical relations.The paper comes into three parts:a) through the analysis of sentence struct ures, a semantic model pool (i.e. a set of semantic models) has been built; b) an algorithm for direct statement of numerical relationships by semantic models has been designed and programmed; c) a set of word problems in direct statement, in which numerical relationships are manually marked, has been used to test the nu merical relationships extracted by semantic models; d) some advices on improvem ent have been provided in the end. |