Entity-Aware Dependency-Based Deep Graph Attention Network for Comparative Preference Classification

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5782–5788 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 5782 Entity-Aware Dependency-Based Deep Graph Attention Network for Comparative Preference Classification Nianzu Ma†, Sahisnu Mazumder†, Hao Wang♮, Bing Liu† †Department of Computer Science, University of Illinois at Chicago, USA ♮School of Information Science and Technology, Southwest Jiaotong University, China jingyima005@gmail.com, sahisnumazumder@gmail.com cshaowang@gmail.com, liub@uic.edu Abstract This paper studies the task of comparative preference classification (CPC). Given two en- tities in a sentence, our goal is to classify whether the first (or the second) entity is pre- ferred over the other or no comparison is ex- pressed at all between the two entities. Ex- isting works either do not learn entity-aware representations well and fail to deal with sen- tences involving multiple entity pairs or use sequential modeling approaches that are un- able to capture long-range dependencies be- tween the entities. Some also use traditional machine learning approaches that do not gener- alize well. This paper proposes a novel Entity- aware Dependency-based Deep Graph Atten- tion Network (ED-GAT) that employs a multi- hop graph attention over a dependency graph sentence representation to leverage both the semantic information from word embeddings and the syntactic information from the depen- dency graph to solve the problem. Empiri- cal evaluation shows that the proposed model achieves the state-of-the-art performance in comparative preference classification. 1 Introduction Given a sentence that contains two entities of in- terest, the task of Comparative Preference Classifi- cation is to decide whether there is a comparison between the two entities and if so, which entity is preferred (Jindal and Liu, 2006a; Ganapathibhotla and Liu, 2008; Liu, 2012; Panchenko et al., 2019). For example, considering sentence s1 (shown in Table 1), there is a comparison between the two underlined entities, and “golf” is preferred over “baseball”. This sentence contains explicit compar- ative predicate “easier”. The task seems straightfor- ward but is quite challenging due to many counter- examples. For example, s2 shows that “better” may not indicate a comparison. s3, another counter- example, shows that “slower” indeed indicates a ID Sentences s1 Golf is easier to pick up than baseball. s2 I’m considering learning Python and more PHP if any of those would be better. s3 The tools based on Perl and Python is much slower under Windows than K9. Table 1: Comparative sentence examples. Entities of interest are underlined in each sentence. comparison, but not between “Perl” and “Python”, but between “tools” and “K9”. Problem statement. Given a sentence s = ⟨w1, w2, ..., e1, ..., e2, ...wn⟩, where e1 and e2 are entities consisting of a single word or a phrase, and e1 appears before e2 in the sentence, our goal is to classify the comparative preference direction between these two entities into one of the three classes: {BETTER, WORSE, NONE}. BETTER (WORSE) means e1 is preferred (not preferred) over e2. NONE means that there is no comparative relation between e1 and e2. Although closely related, Comparative …

Content truncated to 500 words