SetExpan: Corpus-Based Set Expansion via Context Feature Selection and Rank Ensemble

Abstract

Corpus-based set expansion (i.e., finding the "complete" set of entities belonging to the same semantic class, based on a given corpus and a tiny set of seeds) is a critical task in knowledge discovery. It may facilitate numerous downstream applications, such as information extraction, taxonomy induction, question answering, and web search. To discover new entities in an expanded set, previous approaches either make one-time entity ranking based on distributional similarity, or resort to iterative pattern-based bootstrapping. The core challenge for these methods is how to deal with noisy context features derived from free-text corpora, which may lead to entity intrusion and semantic drifting. In this study, we propose a novel framework, SetExpan, which tackles this problem, with two techniques: (1) a context feature selection method that selects clean context features for calculating entity-entity distributional similarity, and (2) a ranking-based unsupervised ensemble method for expanding entity set based on denoised context features. Experiments on three datasets show that SetExpan is robust and outperforms previous state-of-the-art methods in terms of mean average precision.

Guiding Corpus-based Set Expansion by Auxiliary Sets Generation and Co-Expansion

“…Meanwhile, they refine the context feature pool by including only those features which are commonly shared by entities in the expanded set. Based on this philosophy, SetExpan [19] develops a context feature selection module to select quality skip-gram features and designs a rank ensemble module to select quality entities. Similarly, SetExpander [13] captures distributional similarity on five different context types and learns a classifier to combine multiple contexts using an additional labeled dataset.…”

“…Therefore, context dependent similarity benefits set expansion tasks in that it only captures the type-indicative features of entities. We adopt the context dependent similarity function Sim(e i , e j |F ) defined in [19] using the weighted Jaccard similarity measure:…”

“…Then we can update the co-occurrence matrix by where flex(c) means the flexible transformation of local context c. Then we can merge those infrequent but type-indicative skip-grams together. However, one might say a trivial solution to this data sparsity issue is to exhaust a variety of window sizes, (e.g., [-2,+2],[-1,+1],[-1,0],[0,+1]), as has been done by previous studies [18,19,26]. However, this approach very likely ends up generating many skipgrams that are too general.…”

Multi-Context Term Embeddings: the Use Case of Corpus-based Term Set Expansion

“…We focus on corpus-based approaches based on the distributional similarity hypothesis (Harris, 1954). State-of-the-art techniques return the k nearest neighbors around the seed terms as the expanded set, where terms are represented by their co-occurrence or embedding vectors in a training corpus according to different context types, such as linear window context (Pantel et al, 2009;Shi et al, 2010;Rong et al, 2016;Zaheer et al, 2017;Gyllensten and Sahlgren, 2018;Zhao et al, 2018), explicit lists (Roark and Charniak, 1998;Sarmento et al, 2007;He and Xin, 2011), coordinational patterns (Sarmento et al, 2007) and unary patterns (Rong et al, 2016;Shen et al, 2017). In this work, we generalize coordinational patterns, look at additional context types and combine multiple context-type embeddings.…”

Proceedings of the Seventh Joint Conference on Lexical and Computational Semantics

“…Only the n top scoring contexts will have non-zero values in W , and these get the value f ρ . This notion of weighting contexts is similar to that used in the SetExpan framework (Shen et al, 2017), although the way they use it is different (they use weighted Jaccard similarity based on context weights). Their algorithm for calculating context weights is a special case of our algorithm, with no notion of limited support penalty, that is, they use ρ = 0.…”