Multi-Perspective Relevance Matching with Hierarchical ConvNets for Social Media Search

Abstract

Despite substantial interest in applications of neural networks to information retrieval, neural ranking models have only been applied to "standard" ad hoc retrieval tasks over web pages and newswire documents. This paper proposes MP-HCNN (Multi-Perspective Hierarchical Convolutional Neural Network) a novel neural ranking model specifically designed for ranking short social media posts. We identify document length, informal language, and heterogeneous relevance signals as features that distinguish documents in our domain, and present a model specifically designed with these characteristics in mind. Our model uses hierarchical convolutional layers to learn latent semantic soft-match relevance signals at the character, word, and phrase levels. A pooling-based similarity measurement layer integrates evidence from multiple types of matches between the query, the social media post, as well as URLs contained in the post. Extensive experiments using Twitter data from the TREC Microblog Tracks 2011-2014 show that our model significantly outperforms prior feature-based as well and existing neural ranking models. To our best knowledge, this paper presents the first substantial work tackling search over social media posts using neural ranking models.

A Deep Look into neural ranking models for information retrieval

“…MACM [91] builds a CNN over the interaction matrix from η, uses MLP to generate a layer-wise score for each abstraction level of the CNN, and aggregates all the layers' scores for the final relevance estimation. Similar ideas can also be found in MP-HCNN [92] and MultiMatch [93].…”

Critically Examining the "Neural Hype"

“…The next criticism we anticipate concerns our evidence combination method, simple linear interpolation of scores. While there are much more sophisticated approaches to integrating multiple relevance signals, this approach is commonly used [6,16,19,24,26]. In a separate experiment where we explicitly ignored the retrieval scores, effectiveness was significantly lower.…”

Simple Attention-Based Representation Learning for Ranking Short Social Media Posts

“…We compared against two variants of MP-HCNN; MP-HCNN+QL includes a linear interpolation with QL scores. Table 3 shows the effectiveness of all variants of our model, compared against previous results copied from Rao et al (2019). Model 1 illustrates the effectiveness of the basic BiCNN model with a kernel window size of two; combining different window sizes (Kim, 2014) doesn't yield any improvements.…”

“…Results of various models on the TREC Microblog Tracks datasets. Models 5-8 are copied fromRao et al (2019); note that MP-HCNN exploits URL information (+URL). Models with +QL include interpolation with the QL baseline.…”