【摘要】 聊天机器人中的出行消费意图是指用户为了满足出行的需要,通过文本表达出对出行类产品或者服务的购买意愿.识别出用户的消费意图可以进行相应的产品推荐,增强用户体验.传统的消费意图识别主要使用基于模板匹配或者基于人工特征集合的机器学习方法,这类方法费时费力,扩展性不强.本文将出行消费意图识别任务看成一个分类问题,结合深度学习方法识别用户的出行消费意图,该方法不需要人工构造特征集合或匹配模板.具体而言,本文构建了基于卷积的长短期记忆神经网络(Convolutional-LSTM)模型进行出行消费意图识别,首先通过卷积神经网络(CNN)对用户的聊天文本进行特征抽取,随后进行特征组合并送入长短记忆神经网络(LSTM)进行特征表示学习,最后输出分类结果.实验结果表明,在出行消费意图识别任务上,基于Convolutional-LSTM的模型在F值上优于最好的基线方法 2个百分点.更多还原

【Abstract】 Travel consumption intention in chatting robot is the users in order to meet their travel needs, express the willingness to purchase a product or service. Identifying the user’s intent to consume the product can be recommended to enhance the user’s experience. Traditional consumer intention recognition methods are mainly based on template matching or artificial feature sets, which are time consuming, laborious, and hard to extend.In this paper, we regard the travel consumption intention recognition task as a classification problem and combine the deep learning method to identify the intention. The proposed method does not need to construct the feature set or match templates manually. Specifically, this study uses the convolutional long short-term memory neural network(LSTM) model to identify the travel consumption intention. First, the feature extraction is carried out by creating a convolution neural network(CNN) of the user’s chat text, which is then followed by a combination of features. Then, the features are sent to the LSTM to study the characteristics of the feature representation.Finally, the classification results are outputted. Experimental results show that the convolutional-LSTM model is better than the best baseline method by two percentage points on the F-measure.更多还原