Sentiment analysis of multiple implicit features per sentence in consumer review data
With the rise of e-commerce, online consumer reviews have become crucial for consumers' purchasing decisions. Most of the existing research focuses on the detection of explicit features and sentiments in such reviews, thereby ignoring all that is reviewed implicitly. This study builds, in extension of an existing implicit feature algorithm that can only assign one implicit feature to each sentence, a classifier that predicts the presence of multiple implicit features in sentences. The classifier makes its prediction based on a custom score function and a trained threshold. Only if this score exceeds the threshold, we allow for the detection of multiple implicit feature. In this way, we increase the recall while limiting the decrease in precision. In the more realistic scenario, the classifier-based approach improves the F1-score from 62.9% to 64.5% on a restaurant review data set. The precision of the computed sentiment associated with the detected features is 63.9%.
|Keywords||Feature-based sentiment, Implicit features, Product reviews|
|Persistent URL||dx.doi.org/10.3233/978-1-61499-714-6-241, hdl.handle.net/1765/113076|
|Series||Frontiers in Artificial Intelligence and Applications|
Dosoula, N, Griep, R, Den Ridder, R. (Rick), Slangen, R, Van Luijk, R, Schouten, K.I.M, & Frasincar, F. (2016). Sentiment analysis of multiple implicit features per sentence in consumer review data. Frontiers in Artificial Intelligence and Applications. doi:10.3233/978-1-61499-714-6-241