SemEval-2015 Task 12: Aspect Based Sentiment Analysis

Abstract

This paper describes the SemEval 2016 shared task on Aspect Based Sentiment Analysis (ABSA), a continuation of the respective tasks of 2014 and 2015. In its third year, the task provided 19 training and 20 testing datasets for 8 languages and 7 domains, as well as a common evaluation procedure. From these datasets, 25 were for sentence-level and 14 for text-level ABSA; the latter was introduced for the first time as a subtask in SemEval. The task attracted 245 submissions from 29 teams.

Neural Attention Model for Classification of Sentences that Support Promoting/Suppressing Relationship

“…The neural attention model was previously applied to aspect-based sentiment analysis (ABSA) (Yanase et al, 2016), which has some similarity to the evidence classification in that it classifies sentimental polarities towards a subject S given an aspect (corresponding to V) (Pontiki et al, 2015). A limitation of (Yanase et al, 2016) was that the learned attention layer is tightly attached to each S or V and does not generalize for neverencountered subjects/values.…”

Semantic similarity analysis and application in knowledge graphs

“…Therefore, the commonly used word similarity datasets are used to evaluate the proposed semantic similarity method and graphbased IC based on WordNet and DBpedia. Moreover, the semantic similarity methods are evaluated in an aspect category classification task (Pontiki et al, 2015(Pontiki et al, , 2016a in order to evaluate their performance in a real application. This section presents the datasets, implementation, evaluation and provides a brief discussion about the obtained experimental results.…”

“…Based on this observation, we proposed a similarity-based framework for concept classification, in which concept's features are represented by frequent collocated words while feature vectors are constructed by computing semantic similarity between input words and feature words. We demonstrate the similarity-based classification framework in the Aspect Based Sentiment Analysis (ABSA) (Pontiki et al, 2015(Pontiki et al, , 2016a task of aspect category classification by proposing both unsupervised model and supervised model.…”

“…In SemEval 2015 and 2016, the task sentence-level ABSA has defined a subtask so-called Aspect Category Detection, whose aim is to identify every entity E and attribute A pair, towards which an opinion is expressed in the given text (Pontiki et al, 2015). Specifically, given an input sentence such as "The food was delicious", ABSA needs to detect the E and A pair (category=FOOD#QUALITY) for the target word "food" and to estimate its sentiment either positive or negative.…”

“…The N a i = w k ∈f (a i ) count(w k ) denotes the total count of feature words of category a i . The counts of feature words are derived from the annotated datasets from SemEval ABSA (Pontiki et al, 2015(Pontiki et al, , 2016a. We can define…”