Internet Automatic Sales System is be used to assist people in net business. Since part of the talking continent in purchasing is always asked frequently, so the statistic method like TFIDF is applied to deal this issue. Based on the experiment situation, baseline system based on TFIDF achieved very good effect. But the classic algorithm is not enough to reach real application requirement, the Dice method is introduced here. By using ANN linear regression, the coefficient between TFIDF and Dice has been gotten, and the new sentence similarity calculating algorithm improves the baseline systemȁ9;s performance. This paper just aimed at special field and used limited size of corpora. So FAQ (Frequently Asked Questions) is just one part of sales process QA (Question Answering) and other fields need more analysis and research in the future.
展开▼