Since 1980’s

classification standard of Chinese special geographical schemes have been increasingly con- strutted

e.g.Classification and Codes for the National Land Information（GB/T 13923-1992

GB/T13923-2006）

Clas- sification and Codes for the Features 1:500 1:1000 1:2000 Topographic Maps（GB/T14804-1993）

Classification and Codes for the Features 1:5000 1:10000 1:25000 1:50000 1:100000 Topographic Maps（GB/T15660-1995）.However

there is serious semantic heterogeneity between different geographical information classification schemes.Commonly-used intellectual approaches can normally achieve good performance

but problems always arise if there are insufficient time

money and domain experts.To tackle these problems

it is necessary to introduce effective and efficient automatic approa- ches used.Semantic conversion between classification schemes are oriental or directional

i.e.from classification scheme A（source scheme）to classification scheme B（target scheme）is not the same as semantic conversion from classification scheme B（source scheme）to classification scheme A（target scheme）.Theoretically

if every entry in the source scheme can locate a closely relation entry in the target scheme

then the source scheme is fully compatible with the target scheme

and vice versa.With the investigation of string similarity measures for English and Chinese languages

we select a measure taking into account of the characteristics of Chinese language and class names of geographical information classification schemes.On the assumption that the semantic similarity between two entries from different schemes can be measured by the string similarity of their class names

a model consisting of four conversion patterns is defined.These patterns are given a priority according to their representation capability of the semantic content of the corresponding entries.And then an algorithm is developed to conduct automatic semantic conversion of geographical information classification schemes.In this algorithm

data preprocessing is a key step by means of some natural language processing techniques such as syntax processing

punctuation processing

entry splitting and so on in order that every entry can represent complete and distinct content and concept.The other steps aim to implement the conversion model and then to filter optimal semantic relations. The conclusion of semantic relevance between entries of classification schemes is subjective in any real world task.Never- theless

there is no doubt that intellectual approaches can achieve the best performance.Four measures are defined to esti- mate the performance of intellectual and automatic approaches:out of the total number of entries of the source scheme （1）recall ration is the number of unique source entries of the semantic relation created by one automatic or intellectual conversion approach ;（2）precision ration is the number of unique target entries of the semantic relations created by one automatic or intellectual conversion approach ;（3）accuracy ration is the proportion of correct semantic relations out of the total semantic relations created by the automatic approach;（4）matching ration expresses the proportion of overlaps of se- mantic relations generated by the automatic approach from the whole semantic relations created by the intellectual ap- proach.Finally

we carried out experiments on four classification standard Chinese geographical information schemes.The experimental results show that our proposed approach achieve satisfactory performance.