Multimodal sentiment analysis: Difference between revisions

Revision as of 13:43, 15 June 2018

Revision as of 13:50, 15 June 2018

Line 1:

'''Multimodal sentiment analysis''' is a new dimension of the traditional text-based sentiment analysis, which goes beyond the analysis of texts, and includes other [[Modality (human–computer interaction)|modalities]] such as audio and visual data.<ref>{{cite journal |last1=Soleymani |first1=Mohammad |last2=Garcia |first2=David |last3=Jou |first3=Brendan |last4=Schuller |first4=Björn |last5=Chang |first5=Shih-Fu |last6=Pantic |first6=Maja |title=A survey of multimodal sentiment analysis |journal=Image and Vision Computing |date=September 2017 |volume=65 |pages=3–14 |doi=10.1016/j.imavis.2017.08.003}}</ref> It can be bimodal, which includes different combinations of two modalities, or trimodal, which incorporates three modalities.<ref>{{cite journal |last1=Karray |first1=Fakhreddine |last2=Milad |first2=Alemzadeh |last3=Saleh |first3=Jamil Abou |last4=Mo Nours |first4=Arab |title=Human-Computer Interaction: Overview on State of the Art |journal=International Journal on Smart Sensing and Intelligent Systems |date=2008 |url=http://s2is.org/Issues/v1/n1/papers/paper9.pdf}}</ref> With the extensive amount of social media data available online in different forms such as videos and images, the conventional text-based sentiment analysis has evolved into more complex models of multimodal sentiment analysis<ref name="s1">{{cite journal |last1=Poria |first1=Soujanya |last2=Cambria |first2=Erik |last3=Bajpai |first3=Rajiv |last4=Hussain |first4=Amir |title=A review of affective computing: From unimodal analysis to multimodal fusion |journal=Information Fusion |date=September 2017 |volume=37 |pages=98–125 |doi=10.1016/j.inffus.2017.02.003}}</ref>, which can be applied in the development of [[virtual assistant]]s<ref name ="s5">{{cite web |title=Google AI to make phone calls for you |url=https://www.bbc.com/news/technology-44045424 |website=BBC News |accessdate=12 June 2018 |date=8 May 2018}}</ref>, analysis of YouTube movie reviews<ref name="s4">{{cite journal |last1=Wollmer |first1=Martin |last2=Weninger |first2=Felix |last3=Knaup |first3=Tobias |last4=Schuller |first4=Bjorn |last5=Sun |first5=Congkai |last6=Sagae |first6=Kenji |last7=Morency |first7=Louis-Philippe |title=YouTube Movie Reviews: Sentiment Analysis in an Audio-Visual Context |journal=IEEE Intelligent Systems |date=May 2013 |volume=28 |issue=3 |pages=46–53 |doi=10.1109/MIS.2013.34}}</ref>, analysis of news videos<ref>{{cite journal |last1=Pereira |first1=Moisés H. R. |last2=Pádua |first2=Flávio L. C. |last3=Pereira |first3=Adriano C. M. |last4=Benevenuto |first4=Fabrício |last5=Dalip |first5=Daniel H. |title=Fusing Audio, Textual and Visual Features for Sentiment Analysis of News Videos |journal=arXiv:1604.02612 [cs] |date=9 April 2016 |url=http://arxiv.org/abs/1604.02612}}</ref>, and [[emotion recognition]] (sometimes known as [[emotion]] detection) such as [[depression]] monitoring<ref name = "s6">{{cite journal |last1=Zucco |first1=Chiara |last2=Calabrese |first2=Barbara |last3=Cannataro |first3=Mario |title=Sentiment analysis and affective computing for depression monitoring |journal=2017 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) |date=November 2017 |pages=1988–1995 |doi=10.1109/bibm.2017.8217966 |url=http://doi.ieeecomputersociety.org/10.1109/BIBM.2017.8217966 |publisher=IEEE |language=English}}</ref>, among others.

Similar to the traditional sentiment analysis, one of the most basic taskstask in multimodal sentiment analysis is [[sentiment]] classification, which is classifyingclassifies different [[sentiment]]s into categories such as positive, negative, or neutral<ref>{{cite book |last1=Pang |first1=Bo |last2=Lee |first2=Lillian |title=Opinion mining and sentiment analysis |date=2008 |publisher=Now Publishers |___location=Hanover, MA |isbn=1601981503}}</ref>. The complexity of analyzing text, audio, and visual features to perform such a task, requires the application of different fusion techniques, such as feature-level, decision-level, and hybrid fusion.<ref name="s1"></ref> The performance of these fusion techniques and the classification algorithms applied, are influenced by the type of textual, audio, and visual features employed in the analysis.<ref name = "s7"></ref>

== Features ==

Line 7:

=== Textual Features ===

Similar to the conventional text-based sentiment analysis, some of the most commonly used textual features in multimodal sentiment analysis are [[n-grams|unigrams]] and [[n-gram]]s, which are basically a sequence of words in a given textual document.<ref>{{cite journal |last1=Yadollahi |first1=Ali |last2=Shahraki |first2=Ameneh Gholipour |last3=Zaiane |first3=Osmar R. |title=Current State of Text Sentiment Analysis from Opinion to Emotion Mining |journal=ACM Computing Surveys |date=25 May 2017 |volume=50 |issue=2 |pages=1–33 |doi=10.1145/3057270}}</ref> These features are applied using [[bag-of-words]] or bag-of-concepts feature representations, in which words or concepts are represented as vectors in a suitable space.<ref name="s2">{{cite journal |last1=Perez Rosas |first1=Veronica |last2=Mihalcea |first2=Rada |last3=Morency |first3=Louis-Philippe |title=Multimodal Sentiment Analysis of Spanish Online Videos |journal=IEEE Intelligent Systems |date=May 2013 |volume=28 |issue=3 |pages=38–45 |doi=10.1109/MIS.2013.9}}</ref><ref>{{cite journal |last1=Poria |first1=Soujanya |last2=Cambria |first2=Erik |last3=Hussain |first3=Amir |last4=Huang |first4=Guang-Bin |title=Towards an intelligent framework for multimodal affective data analysis |journal=Neural Networks |date=March 2015 |volume=63 |pages=104–116 |doi=10.1016/j.neunet.2014.10.005}}</ref>

=== Audio Features ===

Line 15:

=== Visual Features ===

One of the main advantages of analyzing videos overwith respect to texts alone, is the presence of rich [[sentiment]] cues in visual data.<ref>{{cite journal |last1=Poria |first1=Soujanya |last2=Cambria |first2=Erik |last3=Hazarika |first3=Devamanyu |last4=Majumder |first4=Navonil |last5=Zadeh |first5=Amir |last6=Morency |first6=Louis-Philippe |title=Context-Dependent Sentiment Analysis in User-Generated Videos |journal=Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) |date=2017 |doi=10.18653/v1/p17-1081 |url=https://doi.org/10.18653/v1/P17-1081 |publisher=Association for Computational Linguistics}}</ref> Visual features include [[facial expression]]s, which are principalof signsparamount ofimportance in understandingcapturing [[sentiment]]s and [[emotion]]s, as they are a main channel of forming a person's present state of mind.<ref name="s1"></ref> Specifically, [[smile]], is considered to be one of the most predictive visual cues in multimodal sentiment analysis.<ref name="s2"></ref> OpenFace is an open-source facial analysis toolkit available for extracting and understanding such visual features.<ref>{{cite journal |title=OpenFace: An open source facial behavior analysis toolkit - IEEE Conference Publication |journal=ieeexplore.ieee.org |url=https://ieeexplore.ieee.org/document/7477553/}}</ref>

== Fusion Techniques ==

Unlike the traditional text-based sentiment analysis, multimodal sentiment analysis undergoesundergo a fusion process in which data from different modalities (text, audio, or visual) are fused and analyzed together.<ref name ="s1"></ref> The existing approaches in multimodal sentiment analysis [[data fusion]] can be grouped into three main categories: feature-level, decision-level, and hybrid fusion, and the performance of the [[sentiment]] classification depends on which type of fusion technique is employed.<ref name ="s1"></ref>

=== Feature-level Fusion ===

Feature-level fusion (sometimes known as early fusion), gathers all the features from each [[modality (human–computer interaction)|modality]] (text, audio, or visual) and joins them together into a single feature vector, which is eventually fed into a classification algorithm<ref name="s3">{{cite journal |last1=Poria |first1=Soujanya |last2=Cambria |first2=Erik |last3=Howard |first3=Newton |last4=Huang |first4=Guang-Bin |last5=Hussain |first5=Amir |title=Fusing audio, visual and textual clues for sentiment analysis from multimodal content |journal=Neurocomputing |date=January 2016 |volume=174 |pages=50–59 |doi=10.1016/j.neucom.2015.01.095}}</ref>. One of the difficulties in implementing this technique is the integration of the heterogeneous features.<ref name="s1"></ref>

=== Decision-level Fusion ===