Effectively making sense of short texts is a critical task for many realworld applications such as search engines, social media services, andrecommender systems. The task is particularly challenging as a short textcontains very sparse information, often too sparse for a machine learningalgorithm to pick up useful signals. A common practice for analyzing short textis to first expand it with external information, which is usually harvestedfrom a large collection of longer texts. In literature, short text expansionhas been done with all kinds of heuristics. We propose an end-to-end solutionthat automatically learns how to expand short text to optimize a given learningtask. A novel deep memory network is proposed to automatically find relevantinformation from a collection of longer documents and reformulate the shorttext through a gating mechanism. Using short text classification as ademonstrating task, we show that the deep memory network significantlyoutperforms classical text expansion methods with comprehensive experiments onreal world data sets.
展开▼
