MIX: Multi-Channel Information Crossing for Text Matching

Short Text Matching plays an important role in many natural language processing tasks such as information retrieval, question answering, and dialogue systems. Conventional text matching methods rely on predefined templates and rules. However, for a short piece of text with a limited number of words, these rules are unable to generalize well to unobserved data. With the success of deep learning in fields like computer vision, speech recognition and recommender systems, many recent efforts have been made to apply deep neural network models to natural language processing tasks to reduce the cost of manual feature engineering. In this paper, we present the design of Multi-Channel Information Crossing (MIX), a multi-channel convolutional neural network (CNN) model for text matching in a production environment, with additional attention mechanisms on sentences and semantic features. MIX compares text snippets at varied granularities to form a series of multi-channel similarity matrices, which are then crossed with another set of carefully designed attention matrices to expose the rich structures of sentences to deep neural networks. We implemented MIX and deployed the system on Tencent's Venus distributed computation platform. Thanks to the well-engineered multi-channel information crossing, evaluation results suggest that MIX outperforms a wide range of state-of-the-art deep neural network models by at least 11.1% in terms of the normalized discounted cumulative gain (NDCG@3), on the English WikiQA dataset. Moreover, we performed online A/B tests with real users on the search service of Tencent QQ Browser. Results show that MIX raised the number of clicks on the returned results by 5.7%, due to the increased accuracy in query-document matching, which demonstrates the superior performance of MIX in a real-world production environment.