Mining fine-grained opinions on closed captions of YouTube videos with an attention-RNN

Abstract

Video reviews are the natural evolution of written product reviews. In this paper we target this phenomenon and introduce the first dataset created from closed captions of YouTube product review videos as well as a new attention-RNN model for aspect extraction and joint aspect extraction and sentiment classification. Our model provides state-of-the-art performance on aspect extraction without requiring the usage of hand-crafted features on the SemEval ABSA corpus, while it outperforms the baseline on the joint task. In our dataset, the attention-RNN model outperforms the baseline for both tasks, but we observe important performance drops for all models in comparison to SemEval. These results, as well as further experiments on domain adaptation for aspect extraction, suggest that differences between speech and written text, which have been discussed extensively in the literature, also extend to the domain of product reviews, where they are relevant for fine-grained opinion mining.

Exploiting BERT for Multimodal Target Sentiment Classification through Input Space Translation

“…Xu [40] introduced a Multi-Interactive Memory Network for multimodal aspect sentiment classification, and a private Chinese-language dataset called Multi-ZOL curated from review sites. The YouTubean dataset was introduced by Marrese-Taylor [19], and contains target annotations 500 YouTube review videos.…”

A Multi-modal Approach to Fine-grained Opinion Mining on Video Reviews

“…Finally, Marrese-Taylor et al (2017) and Garcia et al (2019b) contributed multi-modal datasets obtained from product and movie reviews respectively, specifically for the task of fine-grained opinion mining. Furthermore, Garcia et al (2019a) recently used the latter to propose a hierarchical multi-modal model for opinion mining.…”

“…Although reviews often come under the form of a written commentary, people are increasingly turning to video platforms such as YouTube looking for product reviews to help them shop. In this context, Marrese-Taylor et al (2017) explored a new direction, arguing that video reviews are the natural evolution of written product reviews and introduced a dataset of annotated video product review transcripts. Similarly, Garcia et al (2019b) recently presented an improved version of the POM movie review dataset (Park et al, 2014), with annotated fine-grained opinions.…”

“…In light of this, our paper introduces a multi-modal approach for fine-grained opinion mining. We conduct extensive experiments on two datasets built upon transcriptions of video reviews, Youtubean (Marrese-Taylor et al, 2017) and a fine-grain annotated version of the Persuasive Opinion Multimedia (POM) dataset (Park et al, 2014;Garcia et al, 2019b), adapting them to our setting by associating timestamps to each annotated sentence using the video subtitles. Our results demonstrate the effectiveness of our proposed approach and show that by leveraging the additional modalities we can consistently obtain better performance.…”

KT-Speech-Crawler: Automatic Dataset Construction for Speech Recognition from YouTube Videos

“…In addition, YouTube has been used previously in multiple ways to automatically collect multimodal datasets, e.g. emotion recognition datasets by Barros et al (2018) and Zadeh et al (2016), or opinion mining (Marrese-Taylor et al, 2017), or video classification (YouTube-8M 8 , or human action recognition (Kay et al, 2017)).…”