Instance Level Transfer Learning for Cross Lingual Opinion Analysis

摘要

This paper presents two instance-level transfer learning based algorithms for cross lingual opinion analysis by transferring useful translated opinion examples from other languages as the supplementary training data for improving the opinion classifier in target language. Starting from the union of small training data in target language and large translated examples in other languages, the Transfer AdaBoost algorithm is applied to iteratively reduce the influence of low quality translated examples. Alternatively, starting only from the training data in target language, the Transfer Self-training algorithm is designed to iteratively select high quality translated examples to enrich the training data set. These two algorithms are applied to sentence- and document-level cross lingual opinion analysis tasks, respectively. The evaluations show that these algorithms effectively improve the opinion analysis by exploiting small target language training data and large cross lingual training data.