基于黑白名单构建并约束的RNN文本分类

s a fast and efficient data mining technology, text classification has become a research hotspot. Currently, text classification methods based on neural networks have achieved amazing classification performance. However, neural networks are not interpretable and require a large amounts of training data, which are not suitable for cold-start and zero-shot scenarios. On the other hand, rule-based text classification methods are interpretable, require no training data, and often achieve decent classification accuracy. However, rule-based text classification continuously requires to manually update the rule to improve accuracy, which is inefficient. And when the rules are iterated to a certain extent, the accuracy is difficult to improve, resulting underperformmance compared with neural networks in rich-resource scenarios. As a type of rules, blacklist/whitelist have been widely used in spam filtering, news classification and other fields because of its uncomplicated construction. In order to build a text classification model with high accuracy and good interpretability, which can be applied to cold-start, zero-shot, low-resource and rich-resource scenarios, this paper constructs a recurrent neural network based on blacklist/whitelist, and proposes a text classification framework. After that, it is optimized for the problems existing in the process of constructing the recurrent neural network. Finally, this paper improve interpretability of the model by constraining the recurrent neural network with blacklist/whitelist. Experiments on different datasets verifiy the effectiveness of the method proposed in this paper .