Chinese address compound word segmentation technology based on rules and statistic model

Abstract

The invention discloses a Chinese address compound word segmentation technology based on rules and a statistic model. The word segmentation processing is carried out on addresses by comprehensively utilizing a condition random field model and a maximum matching algorithm optimized by the rules; and the condition random field model extracts related characteristics of address inner information, and a training data set established by a pre-processing phase to train the model so that the Chinese address compound word segmentation technology has the capabilities of automatically segmenting address information and identifying address factors. The condition random field model has a strong model identification capability and can be used for successfully identifying cells with database omission, and also has a good ambiguous address identification capability so that the address factors can be successfully distinguished. An MMSEG algorithm has the characteristics of rapid speed, high precision and the like under the good condition of dictionary data support. According to the Chinese address compound word segmentation technology, the two algorithms are combined and can be mutually supplemented and identified, so that the address matching accuracy is effectively improved, and the word segmentation accuracy is improved under the condition that the efficiency is guaranteed.

Description

technical field [0001] The invention belongs to the technical field of geographic information, and in particular relates to a Chinese address compound word segmentation technology based on rules and statistical models. Background technique [0002] Address matching is the process of establishing a correspondence between a literal description address and its geographic location coordinates. The address matching service follows specific steps to find a match for an address. First, the address is standardized; then the server searches address matching reference data to find potential locations; assigns a score to each candidate location based on its proximity to the address, and finally The highest value to match this address. At present, there are many address word segmentation modes in the address matching methods in the prior art, some of which have single word segmentation modes, do not use different technologies in combination, or the combination efficiency of different ...

AI Technical Summary

Problems solved by technology

At present, there are many address word segmentation modes in the address matching methods in the prior art, some of which have a single word segmentation mode, and do not use different technologies in combination, or the combination efficiency of different technologies is not high, resulting in the word segmentation technology based on a single rule. The effective recognition of new words and the slow speed of word segmentation based on a single statistical model, if the two can be used together, improving the accuracy of word segmentation under the premise of ensuring efficiency has become an urgent need