Joint Prediction for Entity/Event-Level Sentiment Analysis using Probabilistic Soft Logic Models

Abstract

In this work, we build an entity/event-level sentiment analysis system, which is able to recognize and infer both explicit and implicit sentiments toward entities and events in the text. We design Probabilistic Soft Logic models that integrate explicit sentiments, inference rules, and +/-effect event information (events that positively or negatively affect entities). The experiments show that the method is able to greatly improve over baseline accuracies in recognizing entity/event-level sentiments.

How Factuality Determines Sentiment Inferences

“…Recently, Deng and Wiebe (2014) and Deng and Wiebe (2015) have introduced an advanced conceptual framework for inferring sentiment implicatures. Their work is most similar to our approach.…”

“…Deng and Wiebe (2015)), we stress the point that verb signatures in the sense of Karttunen (2012) that capture (non-)factuality information regarding complement clauses need to be taken into account in order to properly draw such inferences. We focus on complex sentences where a matrix verb restricts its subclauses with respect to factuality depending on its affirmative status (i.e.…”

How can NLP Tasks Mutually Benefit Sentiment Analysis? A Holistic Approach to Sentiment Analysis

“…The sentiment lexicons are used to recognize Oh no as a negative opinion, the semantic role labeling features are used to recognize the target is the defeating event (Yang and Cardie, 2013), and the implicatures are used to recognize the writer is positive toward the bill since the writer is negative toward the defeating event which harms the bill (Deng et al, 2014;Deng and Wiebe, 2015). These work mainly rely on the clues that directly indicate opinions (e.g., recognizing On no as a negative opinion), or indicate components of opinions (e.g., recognizng the target being defeating), or indicate other opinions based on the information within the sentence (e.g., recognizing a positive opinion toward the bill).…”

“…For example, in (Ex2) a co-reference resolution needs to be run first to infer the writer is positive toward Obama, while in (Ex3) the positive sentiments needs 53 to be recognized first to infer the word they refer to the allies. Previous work (Deng et al, 2014;Deng and Wiebe, 2015) develop joint models to infer sentiments based on the implicature rules (e.g, (Ex1)). They first develop independent systems to recognize sentiments and components of sentiments.…”

“…Most of previous work directly assign scores to pos(X,Y) and neg(X,Y) without any dependency rule. Some recent work (Deng et al, 2014;Deng and Wiebe, 2015) take the scores as local scores and maximize the sum of scores of all the Primary Task and Implicature Knowledge atoms in Table 1 w.r.t. the constraints defined by a subset of implicature rules in (Wiebe and Deng, 2014a).…”