Helpfulness of product reviews as a function of discrete positive and negative emotions

Abstract

The product review plays an important role in customer's purchase decision making process on the e-commerce websites. Emotions can significantly influence the way that reviews are processed. The importance of discrete emotions embedded in online reviews and their impact on review helpfulness is not explored intensively in prior studies. This study builds a helpfulness predictive model using deep neural network and investigates the influences of emotions that contribute to review helpfulness. We present an approach that extract novel discrete positive and negative emotion features from textual content of product reviews using NRC emotion Lexicon. In addition, the type of product, reviewer, visibility, readability, linguistics and sentiment related characteristics are also used for comparison and helpfulness prediction. The experimental results on two real-life datasets demonstrate that positive emotion features are the best predictors when individual category of feature is considered. However, negative emotion features and visibility features have comparable performance. Furthermore, the hybrid set of features with positive emotion features produce the best predictive performance for helpfulness of online reviews. The empirical evaluation finds that Trust, Joy and Anticipation (positive emotions); Anxiety and Sadness (negative emotions) are most influential emotion dimensions and have greater impact on perceived helpfulness. The findings of this study highlight the importance of emotions in online reviews and have significant implications for consumer and ecommerce retailers.

Emotions Trump Facts: The Role of Emotions in on Social Media: A Literature Review

“…Roughly one half of the papers focus on online reviews, and found sentiment to be connected to reviewer popularity and perceived helpfulness. Looking into differentiated emotions revealed that the perceived helpfulness of a review depends on which emotions the review contains [57,106].…”

“…Traditionally, sentiment analysis has measured the positive and negative sentiment of a sentence or longer text, but there are recent examples of using more fine-grained approaches based on emotion categories such as the ones mentioned above (e.g. [57,106]). There are two main methodological approaches.…”

“…Plutchik's wheel of emotions defines basic emotions as well as milder variants of them, and describes how they relate to each other [74] (used in e.g. [18,57]), and Ekkekakis defined a hierarchical structure of the affective domain, combining the idea or core emotions and dimensions [22] (used in e.g. [79]).…”

Emotions Identification by Using Unsupervised Aspect Category Based Sentiment Classification

“…For example "The phone has a too amazing camera quality, it makes the pictures look so realistic". [25] This particular review under sentiment analysis would just hold of being a positive text and if you mine further and use aspect based opinion mining it would still give a thumbs up for the aspect category 'camera quality'. But emotions hold a mandatory part of human nature which needs to be addressed.…”

Deep Learning for Distribution Channels' Management

“…In addition, the type of product, reviewer, visibility, readability, linguistics and sentiment related characteristics are also used for comparison and helpfulness prediction. [8] Machine learning can't do feature engineering. Potentially, there are two types of limitations with machine learning: [9] -An algorithm can only work well on data with the assumption of the training data -with data that has different distribution.…”