1 Introduction
1.1 Opinion mining versus social opinion mining
1.2 Issues and challenges
1.3 Systematic review
2 Research method
2.1 Research questions
-
What are the existing opinion mining approaches which make use of user-generated content obtained from social media platforms?
2.2 Search strategy
-
“opinion mining” and “sentiment analysis”: are both included due to the fact that these key terms are used interchangeably to denote the same field of study (Pang and Lee 2008; Cambria et al. 2013), even though their origins differ and hence do not refer to the same concept (Serrano-Guerrero et al. 2015);
-
“microblog”, “social network” and “Twitter”: the majority of the opinion mining and/or sentiment analysis research and development efforts target these two kinds of social media platforms, in particular the Twitter microblogging service.
2.3 Search application
-
Electronic library search engines have different underlying models, thus not always provide required support for systematic searching;
-
Same set of search terms cannot be used for multiple engines e.g., complex logical combination not supported by the ACM Digital Library but is by the IEEE Xplore Digital Library;
-
Boolean search string is dependent on the order of terms, independent of brackets;
-
Inconsistencies in the order or relevance in search results (e.g., IEEE Xplore Digital Library results are sorted in order of relevance);
-
Certain electronic libraries treat multiple words as a Boolean term and look for instances of all the words together (e.g., “social opinion mining”). In this case, the use of the “AND” Boolean operator (e.g., “social AND opinion AND mining”) looks for all of the words but not necessary together.
Metadata field | ACM | IEEE Xplore | ScienceDirect | SpringerLink |
---|---|---|---|---|
Title | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) |
Abstract | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) |
Keywords | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) |
-
ACM: Separate searches for each metadata field were conducted and results were merged (duplicates removed). Reason being that the metadata field search functionality “ANDs” all metadata fields, whereas manual edition of the search query does not work well when amended.
-
IEEE: Separate searches for each metadata field were conducted and results were merged (duplicates removed).
-
ScienceDirect: One search that takes in consideration all the chosen metadata fields.
-
SpringerLink: By entering a search term or phrase, a search is conducted over the title, abstract and full-text (including authors, affiliations and references) of every article and book chapter. This was noted in the large amount of returned papers (as will be discussed in the next sub-section), which results in a high amount of false positives (and possibly a higher recall).
2.4 Study selection
-
I1. A study that targeted at least one social networking service and/or utilised a social dataset besides other social media services, such as blogs, chats and wikis. Please note that only work performed on social data from social networking services is taken in consideration for the purposes of this review;
-
I2. A study published from the year 2007 onwards. This year was chosen, since the mid-2000s saw the evolution of several social networking services, in particular Facebook’s growth (2007), which currently contains the highest monthly active users;
-
I3. A study published in the English language.
-
E1. A study published before 2007;
-
E2. A study that does not focus on performing any sort of opinion mining on social media services, even though it mentions some of the search terms;
-
E3. A study that focuses on opinion mining or sentiment analysis in general i.e. no reference in a social context;
-
E5. A study that consists of either a paper’s front cover and/or title page.
Primary studies | ACM | IEEE Xplore | ScienceDirect | SpringerLink |
---|---|---|---|---|
Search application | 106 | 242 | 57 | 456 |
False positives | 39 | 83 | 17 | 262 |
Study selection | 67 | 159 | 40 | 194 |
No full paper access | 0 | 0 | 5 | 4 |
Full paper access | 67 | 159 | 35 | 190 |
Total | 451 |
2.5 Extraction of data
2.5.1 Overall
2.5.2 Study selection: electronic libraries
2.5.3 Study selection: additional set
2.6 Synthesis of data
3 Review analysis
Categories | ACM | IEEE Xplore | ScienceDirect | SpringerLink | Additional Set |
---|---|---|---|---|---|
Study selection | 67 | 159 | 40 | 194 | 34 |
No full paper access | 0 | 0 | 5 | 4 | 0 |
Surveys | 2 | 5 | 3 | 8 | 0 |
Work can be applied/used on social data | 1 | 0 | 0 | 1 | 0 |
Organised tasks | 0 | 0 | 0 | 2 | 0 |
3.1 Social media platforms
Social Media Platform | ACM | IEEE Xplore | ScienceDirect | SpringerLink | Additional Set | Total |
---|---|---|---|---|---|---|
Twitter | 53 | 130 | 25 | 136 | 27 | 371 |
Sina Weibo | 4 | 13 | 1 | 26 | 2 | 46 |
Facebook | 4 | 10 | 3 | 10 | 3 | 30 |
YouTube | 7 | 1 | 2 | 1 | 1 | 12 |
Tencent Weibo | 0 | 1 | 1 | 5 | 1 | 8 |
TripAdvisor | 0 | 1 | 2 | 4 | 0 | 7 |
Instagram | 2 | 3 | 0 | 1 | 0 | 6 |
Flickr | 0 | 2 | 0 | 3 | 0 | 5 |
Myspace | 2 | 0 | 0 | 0 | 3 | 5 |
Digg | 2 | 0 | 0 | 0 | 1 | 3 |
Foursquare | 2 | 0 | 0 | 1 | 0 | 3 |
Stocktwits | 0 | 1 | 1 | 0 | 0 | 2 |
LinkedIn | 1 | 0 | 0 | 0 | 0 | 1 |
Plurk | 0 | 0 | 0 | 1 | 0 | 1 |
Weixin | 0 | 0 | 1 | 0 | 0 | 1 |
PatientsLikeMe | 0 | 1 | 0 | 0 | 0 | 1 |
Apontador | 0 | 0 | 0 | 0 | 1 | 1 |
Google+ | 0 | 1 | 0 | 0 | 0 | 1 |
3.2 Techniques
Year | Lx | ML | DL | St | Pr | Fz | Rl | Gr | On | Hy | Mn | Ot |
---|---|---|---|---|---|---|---|---|---|---|---|---|
2007 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
2008 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
2009 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |
2010 | 2 | 2 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 3 | 0 | 0 |
2011 | 2 | 3 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 7 | 1 | 0 |
2012 | 6 | 5 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 10 | 0 | 1 |
2013 | 6 | 14 | 2 | 1 | 1 | 0 | 2 | 0 | 1 | 21 | 0 | 0 |
2014 | 14 | 20 | 2 | 1 | 3 | 1 | 1 | 0 | 1 | 41 | 0 | 3 |
2015 | 16 | 15 | 4 | 1 | 1 | 0 | 0 | 1 | 0 | 42 | 0 | 0 |
2016 | 13 | 21 | 4 | 3 | 0 | 0 | 0 | 0 | 0 | 38 | 2 | 4 |
2017 | 20 | 22 | 9 | 2 | 1 | 1 | 0 | 0 | 0 | 50 | 2 | 5 |
2018 | 17 | 18 | 13 | 1 | 0 | 0 | 1 | 2 | 0 | 69 | 1 | 4 |
Total | 96 | 121 | 35 | 9 | 6 | 2 | 4 | 4 | 2 | 282 | 6 | 17 |
3.2.1 Lexicon
Number of lexicons | 1 | 2 | 3 | 4 | 6 | 7 | 8 | Other/NA |
Number of studies | 39 | 19 | 10 | 4 | 1 | 1 | 1 | 21 |
References of studies |
Rathan et al. (2018), Li and Fleyeh (2018), Geetha et al. (2018), Chen (2018), Aoudi and Malik (2018), Poortvliet and Wang (2018), Salari et al. (2018), Hubert et al. (2018), Ray and Chakrabarti (2017), Gupta and Joshi (2017), Zhang et al. (2017), Arslan et al. (2017), Hagge et al. (2017), Ozer et al. (2017), Ishikawa and Sakurai (2017), Gallegos et al. (2016), Polymerou et al. (2014), Paltoglou and Thelwall (2012), Santarcangelo et al. (2015), Jurek et al. (2014), Philander and YunYing (2016), Walha et al. (2016), Pääkkönen (2016), Gao et al. (2016), Munezero et al. (2015), Hridoy et al. (2015), Jiang et al. (2015), Wang and Wu (2015), Hagen et al. (2015), Lu (2015), Lu et al. (2015), Varshney and Gupta (2014), Ou et al. (2014), Mostafa (2013a), Mostafa (2013b), Lek and Poo (2013), Tumasjan et al. (2010), Raja and Swamynathan (2016) and Lau et al. (2014) |
Singh et al. (2018), Pollacci et al. (2017), Abdullah and Hadzikadic (2017), Joyce and Deng (2017), Husnain et al. (2017), Shi et al. (2017), Khuc et al. (2012), Porshnev et al. (2013), Li et al. (2017), Mukherjee and Mukherjee (2017), Chou et al. (2017), Frankenstein et al. (2016), Giachanou et al. (2016), Feng et al. (2015), Hagen et al. (2015), Del Bosque and Garza (2014), Zhou et al. (2014), Souza and Vieira (2012) and Asiaee et al. (2012) |
Gonçalves et al. (2013) |
Årup Nielsen (2011) |
Erdmann et al. (2014) |
Rathan et al. (2018), Ranjan et al. (2018), Nausheen and Begum (2018), Wang et al. (2018), Vo et al. (2017), Yan et al. (2017), Radhika and Sankar (2017), Hu et al. (2017), Kaushik and Dey (2016), Akcora et al. (2010), Gupta and Kohli (2016), Lai et al. (2015), Dasgupta et al. (2015), Sarlan et al. (2014), Parthasarathi et al. (2012), Tian et al. (2015), Song et al. (2015), Choi and Kim (2013), Park et al. (2011), Costa et al. (2014) and Yanmei and Yuda (2015) |
3.2.2 Machine learning
Number of machine learning algorithms | 1 | 2 | 3 | 4 | 5 | 6 | 7 | NA |
Number of studies | 59 | 23 | 18 | 9 | 5 | 2 | 1 | 4 |
References of studies |
Fatyanosa et al. (2018), dos Santos et al. (2018), Rout et al. (2018), Liu et al. (2018), Katz et al. (2018), Huang et al. (2018), Halibas et al. (2018), Ignatov and Ignatov (2017), Vo et al. (2017), Omar et al. (2017), Ducange and Fazzolari (2017), Joyce and Deng (2017), Soni et al. (2017), Radhika and Sankar (2017), Wehrmann et al. (2017), Song and Gruzd (2017), Li et al. (2017), Sygkounas et al. (2016), Kumar and Bala (2016), Balaji et al. (2016), Severyn et al. (2016), Singh and Kumari (2016), Abdelrazeq et al. (2016), Wang et al. (2016), Nagiwale and Umale (2015), Smailović et al. (2015), Attigeri et al. (2015), D’Avanzo and Pilato (2015), Kokkinogenis et al. (2015), Seron et al. (2015), Lu (2015), Wagner et al. (2015), Liu et al. (2015), Sluban et al. (2015), Lu et al. (2015), Guerra et al. (2014), Du et al. (2014), Yan et al. (2014), Lau et al. (2014), Batista and Ratté (2014), Rao and Srivastava (2014), Tapia and Velásquez (2014), Molina-González et al. (2014), Abdul-Mageed et al. (2014), Li et al. (2014), Ghiassi et al. (2013), Porshnev et al. (2013), Hoang et al. (2013), Kranjc et al. (2013), Gonçalves et al. (2013), Wunnasri et al. (2013), Yu et al. (2013), Weiss et al. (2013), Xiong et al. (2013), Wang et al. (2012), Xie et al. (2012), Saif et al. (2012), Bifet et al. (2011), Bollen et al. (2011) | Zhang et al. 2018, Moh et al. (2017), Nugroho et al. (2017), Gupta and Singal (2017), Shi et al. (2017), Hao et al. (2017), Wang et al. (2016), Suresh (2016), Peng et al. (2016), Ramteke et al. (2016), Shyamasundar and Rani (2016), de Souza Carvalho et al. (2016), Lu et al. (2016), Zhang et al. (2015), Esiyok and Albayrak (2015), Wang et al. (2014), Filice et al. (2014), Garg and Chatterjee (2014), Ou et al. (2014), Politopoulou and Maragoudakis (2013), Mejova et al. (2013), Wang and Ye (2013) and Li and Li (2013) |
Ismail et al. (2018), Baltas et al. (2017), Yan et al. (2017), Vora and Chacko (2017), Sun et al. (2017), Balikas et al. (2017), Anastasia and Budi (2016), Li et al. (2016), Khalil et al. (2015), Anjaria and Guddeti (2014), Zimmermann et al. (2014), Le et al. (2014), Neethu and Rajasree (2013), Paltoglou and Thelwall (2012), Zhang et al. (2011), Bifet and Frank (2010), Pak and Paroubek (2010), Go et al. (2009) |
Juneja and Ojha (2017) |
Algorithm | Number of studies | References |
---|---|---|
Naïve Bayes (NB) | 75 |
Lewis (1998) |
Support Vector Machine (SVM) | 71 |
Cortes and Vapnik (1995) |
Logistic Regression (LoR) | 16 |
McCullagh (1984) |
Decision Tree (DT) | 15 |
Quinlan (1986) |
Maximum Entropy (MaxEnt) | 12 |
Jaynes (1957) |
Random Forest (RF) | 9 |
Breiman (2001) |
K-Nearest Neighbors (KNN) | 7 |
Altman (1992) |
SentiStrength | 5 |
Thelwall et al. (2012) |
Conditional Random Field | 4 |
Lafferty et al. (2001) |
Linear Regression (LiR) | 4 |
Cook (1977) |
SANT optimization algorithm (SANT) | 3 |
Hu et al. (2013) |
Stochastic Gradient Descent (SGD) | 3 |
Bottou (2010) |
Passive Aggressive (PA) | 2 |
Crammer et al. (2006) |
Bootstrap Aggregating (Bagging) | 1 |
Breiman (1996) |
Bayesian Network (BN) | 1 |
Heckerman et al. (1995) |
Conjunctive Rule Based (CRB) | 1 |
Clark and Niblett (1989) |
Adaptive Boosting (AB) | 1 |
Freund et al. (1999) |
Hidden Markov Model (HMM) | 1 |
Baum and Petrie (1966) |
Dictionary Learning | 1 |
Ramirez et al. (2010) |
SVM with NB features (NBSVM) | 1 |
Wang and Manning (2012) |
Multiclass Classifier (MCC) | 1 |
Witten et al. (2016) |
Iterative Classifier Optimizer (ICO) | 1 |
Witten et al. (2016) |
-
Çeliktuğ (2018) used two ensemble learning methods in RF and MCC (amongst other machine learning algorithms), for sentiment classification of Twitter datasets;
-
Yan et al. (2017) presented two ensemble learners built on four off-the-shelf classifiers, for Twitter sentiment classification;
-
Troussas et al. (2016) evaluated the most common ensemble methods that can be used for sentiment analysis on Twitter datasets;
-
Sygkounas et al. (2016) proposed an ensemble system composed on five state-of-the-art sentiment classifiers;
-
Le et al. (2014) used multiple oblique decision stumps classifiers to form an ensemble of classifiers, which is more accurate than a single one for classifying tweets;
-
Neethu and Rajasree (2013) used an ensemble classifier (and single algorithm classifiers) for sentiment classification.
3.2.3 Deep learning
Algorithm | Number of studies | References |
---|---|---|
Long Short-Term Memory (LSTM) | 13 |
Hinton et al. (2012) |
Convolutional Neural Network (CNN) | 12 |
LeCun et al. (1990) |
Recurrent Neural Network (RNN) | 8 |
Graves and Schmidhuber (2005) |
ANN | 5 |
McCulloch and Pitts (1943) |
Recursive Neural Tensor Network (RNTN) | 3 |
Socher et al. (2013) |
Bidirectional Long Short-Term Memory (BLSTM) | 3 |
Graves and Schmidhuber (2005) |
Multilayer Perceptron (MLP) | 2 |
Hornik et al. (1989) |
Autoencoder (AE) | 2 |
Rumelhart et al. (1985) |
Gated Recurrent Units (GRU) | 1 |
Greff et al. (2017) |
Dynamic Architecture for ANN (DAN2) | 1 |
Ghiassi and Saidane (2005) |
3.2.4 Statistical
3.2.5 Probabilistic
3.2.6 Fuzziness
3.2.7 Rule-based
3.2.8 Graph
3.2.9 Ontology
3.2.10 Hybrid
Lx | ML | DL | St | Pr | Fz | Rl | Gr | On | Total | Studies |
---|---|---|---|---|---|---|---|---|---|---|
✓ | ✓ | 114 |
Zhang et al. (2018), Yan et al. (2018), Pollacci et al. (2017), Jin et al. (2017), Hong and Sinnott (2018), Calvo and Juárez Gambino (2018), Rathan et al. (2018), Saleena (2018), Yan et al. (2017), Katiyar et al. (2018), Gandhe et al. (2018), Al Shammari (2018), Pai and Liu (2018), Goel et al. (2018), Sahni et al. (2017), Ahuja and Dubey (2017), Fatyanosa and Bachtiar (2017), Singh et al. (2018), Abdullah and Zolkepli (2017), Lee and Nerghes (2017), Bouchlaghem et al. (2016), Sharma et al. (2016), Sankaranarayanan et al. (2016), Kanavos et al. (2016), Gallegos et al. (2016), Koto and Adriani (2015), Buscaldi and Hernandez-Farias (2015), Tsytsarau et al. (2014), Yuan et al. (2014), Bravo-Marquez et al. (2013), Zhang et al. (2013), Xu et al. (2012), Jianqiang and Xiaolin (2017), Qaisi and Aljarah (2016), Zimbra et al. (2016), Jianqiang (2016), You and Tunçer (2016), Bravo-Marquez et al. (2016), Zhao et al. (2016), Li et al. (2016), Deshwal and Sharma (2016), Jianqiang and Xueliang (2015), Chen et al. (2015), Li et al. (2016), Fersini et al. (2015), Abdelwahab et al. (2015), Yang et al. (2015), Yang and Zhou (2015), Chen et al. (2015), Kanakaraj and Guddeti (2015), Jianqiang (2015), Koto and Adriani (2015), Wu et al. (2015), Shukri et al. (2015), Sahu et al. (2015), Lewenberg et al. (2015), Cho et al. (2014), Sui et al. (2012), Karyotis et al. (2017), Lim et al. (2017), Pandey et al. (2017), Burnap et al. (2016), Lima et al. (2015), Poria et al. (2014), Bravo-Marquez et al. (2014), Da Silva et al. (2014), Gambino and Calvo (2016), Yan and Tao (2016), Jiang et al. (2015), Nguyen (2016), Aboluwarin et al. (2016), Zainuddin et al. (2016), Flaes et al. (2016), Koto and Adriani (2015), Hagen et al. (2015), Castellucci et al. (2015), Sanborn et al. (2015), Castellucci et al. (2015), Chen et al. (2015), Mansour et al. (2015), Del Bosque and Garza (2014), Han and Kavuluru (2015), Yuan et al. (2015), Buddhitha and Inkpen (2015), Ji et al. (2015), Zhou et al. (2014), Wang et al. (2014), Tsakalidis et al. (2014), Porshnev and Redkin (2014), Su et al. (2014), Yan et al. (2014), Gonçalves et al. (2013), Porshnev et al. (2014), Sun et al. (2014), Pla and Hurtado (2014), Wang and Li (2014), Bao et al. (2014), Zhu et al. (2013), Jiang et al. (2013), Cui et al. (2013), Khuc et al. (2012), Bermingham and Smeaton (2010), Wang et al. (2012), Montejo-Raez et al. (2014), Ortigosa et al. (2014), Rui et al. (2013), Reyes et al. (2013), Kouloumpis et al. (2011), Bakliwal et al. (2013), Vu et al. (2012), Agarwal et al. (2011), Hernandez-Farias et al. (2014), Thelwall et al. (2011) and Thelwall et al. (2010) | |||||||
✓ | ✓ | 12 | ||||||||
✓ | ✓ | 22 |
Zhang et al. (2018), Wan et al. (2018), Sangameswar et al. (2017), Rout et al. (2018), Rout et al. (2017), Bansal and Srivastava (2018), Satapathy et al. (2017), Tago and Jin (2018), Fatyanosa and Bachtiar (2017), Sachdeva et al. (2018), Zhou et al. (2017), Le et al. (2017), Azzouza et al. (2017), Giachanou and Crestani (2016), Gao et al. (2016), Lu et al. (2016), Orellana-Rodriguez et al. (2015), Tan et al. (2014), Khan et al. (2014), Orellana-Rodriguez et al. (2013), Blenn et al. (2012) and Zhang et al. (2012) | |||||||
✓ | ✓ | 3 | ||||||||
✓ | ✓ | 4 | ||||||||
✓ | ✓ | 9 | ||||||||
✓ | ✓ | 4 | ||||||||
✓ | ✓ | 2 | ||||||||
✓ | ✓ | 7 | ||||||||
✓ | ✓ | 21 |
Wang et al. (2018), Saidani et al. (2017), Sabuj et al. (2017), Ismail et al. (2018), Hanafy et al. (2018), Elouardighi et al. (2017), Effrosynidis et al. (2017), Ameur et al. (2018), Symeonidis et al. (2018), Rezaei and Jalali (2017), Elzayady et al. (2018), Rinaldi and Musdholifah (2017), Coyne et al. (2017), Setiawan et al. (2018), Dedhia and Ramteke (2017), Alzahrani (2018), Elbagir and Yang (2018), Mishra and Diesner (2018), Ramadhani et al. (2016), Trung et al. (2013) and Taddy (2013) | |||||||
✓ | ✓ | 3 | ||||||||
✓ | ✓ | 2 | ||||||||
✓ | ✓ | 3 | ||||||||
✓ | ✓ | 3 | ||||||||
✓ | ✓ | 1 |
Haldenwang et al. (2018) | |||||||
✓ | ✓ | 1 |
Mukkamala et al. (2014) | |||||||
✓ | ✓ | 1 |
Karyotis et al. (2017) | |||||||
✓ | ✓ | 1 |
Mukkamala et al. (2014) |
Lx | ML | DL | St | Pr | Fz | Rl | Gr | On | Total | Studies |
---|---|---|---|---|---|---|---|---|---|---|
✓ | ✓ | ✓ | 3 | |||||||
✓ | ✓ | ✓ | 21 |
Vo et al. (2017), Villegas et al. (2018), Konate and Du (2018), Giachanou et al. (2017), Alharbi and DeDoncker (2017), Saleena (2018), Ghiassi and Lee (2018), Tellez et al. (2017), Simões et al. (2017), Lavanya and Deisy (2017), Li and Fleyeh (2018), Permatasari et al. (2018), Fitri et al. (2018), Bouazizi and Ohtsuki (2018), Rai et al. (2018), Wijayanti and Arisal (2017), Xia et al. (2017), Jianqiang et al. (2018), Bouazizi and Ohtsuki (2017), Chen et al. (2015) and Pei et al. (2014) | ||||||
✓ | ✓ | ✓ | 2 | |||||||
✓ | ✓ | ✓ | 12 | |||||||
✓ | ✓ | ✓ | 7 | |||||||
✓ | ✓ | ✓ | 2 | |||||||
✓ | ✓ | ✓ | 2 | |||||||
✓ | ✓ | ✓ | 1 |
Ji et al. (2016) | ||||||
✓ | ✓ | ✓ | 1 |
Zhang et al. (2018) | ||||||
✓ | ✓ | ✓ | 1 |
Dragoni (2018) | ||||||
✓ | ✓ | ✓ | 3 | |||||||
✓ | ✓ | ✓ | 2 | |||||||
✓ | ✓ | ✓ | 1 |
Tong et al. (2017) | ||||||
✓ | ✓ | ✓ | 1 |
Samoylov (2014) | ||||||
✓ | ✓ | ✓ | 2 | |||||||
✓ | ✓ | ✓ | 1 |
Nivetha et al. (2016) | ||||||
✓ | ✓ | ✓ | ✓ | 2 | ||||||
✓ | ✓ | ✓ | ✓ | 1 |
Vo et al. (2017) | |||||
✓ | ✓ | ✓ | ✓ | 3 | ||||||
✓ | ✓ | ✓ | ✓ | 1 |
Kuo et al. (2016) |
Algorithm | Learning type | Studies # |
---|---|---|
SVM | Sup | 130 |
NB | Sup | 96 |
LoR | Sup | 34 |
DT | Sup | 27 |
RF | Sup | 21 |
MaxEnt | Sup | 15 |
SentiStrength | Sup/Unsup | 13 |
LiR | Sup | 8 |
KNN | Sup | 5 |
AB | Sup | 5 |
BN | Sup | 3 |
Support Vector Regression (SVR) | Sup | 3 |
SANT | Sup | 3 |
KM | Unsup | 2 |
Repeated Incremental Pruning to Produce Error Reduction (RIPPER) | Sup | 2 |
HMM | Sup | 1 |
Extremely Randomised Trees | Sup | 1 |
Least Median of Squares Regression | Sup | 1 |
Maximum Likelihood Estimation | Sup | 1 |
Hyperpipes | Sup | 1 |
Extreme Learning Machine | Sup | 1 |
Domain Adaptation Machine | Sup | 1 |
Affinity Propagation | Unsup | 1 |
Multinomial inverse regression | Unsup | 1 |
Apriori | Sup/Unsup | 1 |
Distant Supervision | Semi-sup | 1 |
Label Propagation | Semi-sup | 1 |
SGD | Sup | 1 |
NBSVM | Sup | 1 |
-
CNN—used in 16 studies (Yan et al. 2018; Stojanovski et al. 2018; Konate and Du 2018; Hanafy et al. 2018; Haldenwang et al. 2018; Ghosal et al. 2018; Chen et al. 2017; Ameur et al. 2018; Alharbi and DeDoncker 2017; Symeonidis et al. 2018; Saini et al. 2018; Jianqiang et al. 2018; Baccouche et al. 2018; Cai and Xia 2015; Kalayeh et al. 2015; Yanmei and Yuda 2015);
-
GRU—used in 1 study (Cao et al. 2018);
3.2.11 Other
3.3 Social datasets
3.3.1 Overview
3.3.2 Comparative analysis
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 87.60% (Jianqiang et al. 2018) | 91.76% (Siddiqua et al. 2016) | 87.50% (Jianqiang et al. 2018) | 89.61% (Shyamasundar and Rani 2016) |
2 | 85.00% (Ismail et al. 2018) | 87.60% (Bravo-Marquez et al. 2013) | 86.08% (Siddiqua et al. 2016) | 88.80% (Arslan et al. 2018) |
3 | 84.56% (Zainuddin et al. 2016) | 87.45% (Jianqiang et al. 2018) | 83.90% (Saif et al. 2012) | 88.30% (Lek and Poo 2013) |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 97.72% (Ameur et al. 2018) | 97.41% (Ameur et al. 2018) | 98.20% (Bravo-Marquez et al. 2014) | 98.10% (Bravo-Marquez et al. 2014) |
2 | 79.00% (Bravo-Marquez et al. 2013) | 89.10% (Bravo-Marquez et al. 2013) | 97.57% (Ameur et al. 2018) | 88.93% (Korenek and Šimko 2014) |
3 | 77.60% (Deshwal and Sharma 2016) | 78.70% (Deshwal and Sharma 2016) | 84.85% (Da Silva et al. 2014) | 88.30% (Çeliktuğ 2018) |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 80.69% (Chikersal et al. 2015) | 83.68% (Chikersal et al. 2015) | 93.70% (Bravo-Marquez et al. 2014) | 93.70% (Bravo-Marquez et al. 2014) |
2 | NA | NA | 81.90% (Chikersal et al. 2015) | 91.16% (Lima et al. 2015) |
3 | NA | NA | 80.30% (Xia et al. 2017) | 89.00% (Yan et al. 2018) |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 83.56% (Jianqiang et al. 2018) | 81.48% (Jianqiang et al. 2018) | 82.36% (Xia et al. 2017) | 85.82% (Jianqiang et al. 2018) |
2 | 80.47% (Jianqiang 2016) | 80.98% (Chikersal et al. 2015) | 81.99% (Jianqiang et al. 2018) | 83.82% (Jianqiang 2016) |
3 | 78.93% (Chikersal et al. 2015) | 76.89% (Jianqiang 2016) | 79.81% (Chikersal et al. 2015) | 83.06% (Jianqiang and Xueliang 2015) |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 82.75% (Jianqiang et al. 2018) | 82.61% (Jianqiang et al. 2018) | 83.10% (Saif et al. 2014) | 92.67% (Krouska et al. 2016) |
2 | 82.20% (Ismail et al. 2018) | 82.30% (Ismail et al. 2018) | 82.65% (Jianqiang et al. 2018) | 89.02% (Troussas et al. 2016) |
3 | 79.26% (Saif et al. 2014) | 80.04% (Saif et al. 2014) | 79.62% (Saif et al. 2014) | 86.37% (Yan and Tao 2016) |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 71.30% (Mishra and Diesner 2018) | 67.40% (Saif et al. 2012) | 70.28% (Saleena 2018) | 91.94% (Krouska et al. 2016) |
2 | 69.15% (Saif et al. 2012) | 59.47% (Saif et al. 2014) | 69.10% (Saif et al. 2014) | 85.10% (Troussas et al. 2016) |
3 | 59.76% (Saif et al. 2014) | 58.30% (Mishra and Diesner 2018) | 68.00% (Da Silva et al. 2014) | 84.50% (Yan et al. 2018) |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 81.36% (Zhang et al. 2018) | 79.00% (Saif et al. 2012) | 81.34% (Saif et al. 2014) | 92.59% (Krouska et al. 2016) |
2 | 78.95% (Saif et al. 2012) | 65.76% (Saif et al. 2014) | 78.20% (Saif et al. 2012) | 87.74% (Troussas et al. 2016) |
3 | 66.51% (Saif et al. 2014) | 61.60% (Mishra and Diesner 2018) | 74.65% (Da Silva et al. 2014) | 82.90% (Saif et al. 2014) |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | NA | NA | 76.02% (Xia et al. 2017) | 68.77% (Stojanovski et al. 2015) |
2 | NA | NA | 67.39% (Sygkounas et al. 2016) | 68.00% (Li et al. 2017) |
3 | NA | NA | 64.88% (Stojanovski et al. 2018) | 51.95% (Stojanovski et al. 2018) |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 80.61% (Jianqiang et al. 2018) | 80.86% (Jianqiang et al. 2018) | 80.72% (Jianqiang et al. 2018) | 89.10% (Yan et al. 2018) |
2 | 67.77% (Zhang et al. 2018) | 54.77% (Zhang et al. 2018) | 72.27% (Saif et al. 2014) | 84.59% (Lima et al. 2015) |
3 | NA | NA | 59.27% (Zhang et al. 2018) | 81.56% (Su and Wang 2017) |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 64.10% (Mishra and Diesner 2018) | 60.50% (Mishra and Diesner 2018) | 77.25% (Xia et al. 2017) | 65.60% (Mishra and Diesner 2018) |
2 | NA | NA | 61.40% (Mishra and Diesner 2018) | NA |
3 | NA | NA | 57.10% (Villegas et al. 2018) | NA |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 72.20% (Cui et al. 2013) | 96.70% (Feng et al. 2015) | 78.80% (Feng et al. 2015) | NA |
2 | 69.15% (Hao et al. 2017) | 96.00% (Shi et al. 2013) | 77.00% (Shi et al. 2013) | NA |
3 | 66.90% (Feng et al. 2015) | 73.80% (Cui et al. 2013) | 72.10% (Cui et al. 2013) | NA |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 80.90% (Shi et al. 2013) | 77.80% (Shi et al. 2013) | 79.30% (Shi et al. 2013) | NA |
2 | 78.60% (Cui et al. 2013) | 74.60% (Feng et al. 2015) | 69.14% (Hao et al. 2017) | NA |
3 | 70.83% (Hao et al. 2017) | 67.52% (Hao et al. 2017) | 67.10% (Cui et al. 2013) | NA |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | NA | NA | 83.02% (Xia et al. 2017) | 78.80% (Jiang et al. 2015) |
2 | NA | NA | NA | 63.90% (Jiang et al. 2013) |
3 | NA | NA | NA | NA |
Ranking | Precision | Recall | F-measure | Accuracy |
---|---|---|---|---|
1 | 88.00% (Jianqiang et al. 2018) | 87.32% (Jianqiang et al. 2018) | 87.66% (Jianqiang et al. 2018) | 87.39% (Jianqiang et al. 2018) |
2 | 86.16% (Jianqiang 2016) | 86.15% (Jianqiang 2016) | 86.08% (Jianqiang 2016) | 86.72% (Jianqiang 2016) |
3 | NA | NA | NA | 85.87% (Jianqiang and Xueliang 2015) |
-
In cases where several techniques and/or methods were applied, the highest result obtained in the study for each of the four evaluation metrics, was recorded, even if the technique did not produce the best result for all metrics.
-
The average Precision, Recall, and F-measure results are quoted (if provided by authors), i.e., average score of the results for each classified level (e.g., the average score of the results obtained for each sentiment polarity classification level - positive, negative and, neutral).
-
Results for social datasets that were released as a shared evaluation task, such as SemEval, were either only provided in the metrics used by the task organisers or other metrics were chosen by the authors, therefore not quoted.
-
Certain studies evaluated their techniques based on a subset of the actual dataset. Results quoted are the ones where the entire dataset was used (according to the authors and/our our understanding).
-
Quoted results are for classification tasks and not aspect-based SOM, which can vary depending on the focus of the study.
-
Results presented in a graph visualisation were not considered due to the exact values not being clear.
3.4 Language
-
English and Brazilian Portuguese (Guerra et al. 2014);
-
English and Dutch (Flaes et al. 2016);
-
English and Greek (Politopoulou and Maragoudakis 2013);
-
English and Hindi (Anjaria and Guddeti 2014);
-
English and Japanese (Ragavi and Usharani 2014);
-
English and Roman-Urdu (Javed et al. 2014);
-
English and Swedish (Li and Fleyeh 2018);
-
English and Korean (Ramadhani and Goo 2017);
-
English, German and Spanish (Boididou et al. 2018).
Language | Total | Studies |
---|---|---|
Chinese | 53 |
Cao et al. (2018), Li et al. (2018), Liu et al. (2018), Sun et al. (2018), Wang et al. (2018), Wan et al. (2018), Chou et al. (2017), Hao et al. (2017), Ouyang et al. (2017), Shi et al. (2017), Sun et al. (2017), Zhang et al. (2017), Zhang et al. (2013), Gao et al. (2016), Liu and Young (2016), Zhao et al. (2016), Wang et al. (2016), Wu et al. (2016), Li et al. (2016), Yang and Zhou (2015), Chen et al. (2015), Wang et al. (2014), Sui et al. (2012), Yanmei and Yuda (2015), Liu et al. (2015), Zhang et al. (2015), Wang et al. (2014), Tian et al. (2015), Feng et al. (2015), Song et al. (2015), Jiang et al. (2015), Kuo et al. (2016), Wang et al. (2016), Wang and Wu (2015), Du et al. (2014), Gao et al. (2015), Chen et al. (2015), Wang et al. (2014), Su et al. (2014), Ou et al. (2014), Yan et al. (2014), Pei et al. (2014), Sun et al. (2014), Wang and Li (2014), Xiong et al. (2013), Zhu et al. (2013), Jiang et al. (2013), Zhang et al. (2013), Tang et al. (2013), Cui et al. (2013), Shi et al. (2013), Zhang et al. (2012) and Li and Xu (2014) |
Spanish | 11 |
Calvo and Juárez Gambino (2018), Hubert et al. (2018), Ochoa-Luna and Ari (2018), Sánchez-Holgado and Arcila-Calderón (2018), Gonzalez-Marron et al. (2017), Tellez et al. (2017), Gambino and Calvo (2016), Tapia and Velásquez (2014), Molina-González et al. (2014), Pla and Hurtado (2014) and Ortigosa et al. (2014) |
Indonesian | 8 | |
Italian | 5 | |
Arabic | 5 | |
Portuguese | 3 | |
Brazilian Portuguese | 3 | |
Japanese | 3 | |
Korean | 2 | |
French | 2 | |
French—Bambara | 1 |
Konate and Du (2018) |
Bulgarian | 1 |
Smailović et al. (2015) |
German | 1 |
Rill et al. (2014) |
Roman Urdu | 1 |
Zafar et al. (2016) |
Russian | 1 |
Averchenkov et al. (2015) |
Swiss German | 1 |
Cvijikj and Michahelles (2011) |
Thai | 1 |
Wunnasri et al. (2013) |
Persian | 1 |
Salari et al. (2018) |
Bengala | 1 |
Sabuj et al. (2017) |
Vietnamese | 1 |
Vo et al. (2017) |
3.5 Modality
3.5.1 Datasets
3.5.2 Observations
3.6 Tools and technologies
3.6.1 NLP
-
Natural Language Toolkit (NLTK)75: a platform that provides lexical resources, text processing libraries for classification, tokenisation, stemming, tagging, parsing, and semantic reasoning, and wrappers for industrial NLP libraries;
-
TweetNLP76: consists of a tokeniser, Part-of-Speech (POS) tagger, hierarchical word clusters, and a dependency parser for tweets, besides annotated corpora and web-based annotation tools;
-
Stanford NLP77: software that provides statistical NLP, deep learning NLP and rule-based NLP tools, such as Stanford CoreNLP, Stanford Parser, Stanford POS Tagger;
-
NLPIR-ICTCLAS78: a Chinese word segmentation system that includes keyword extraction, POS tagging, NER, and microblog analysis, amongst other features;
-
word2vec79: an efficient implementation of the continuous bag-of-words and skip-gram architectures for computing vector representations of words.
3.6.2 Machine learning
-
Weka80: a collection of machine learning algorithms for data mining tasks, including tools for data preparation, classification, regression, clustering, association rules mining and visualisation;
-
scikit-learn81: consists of a set of tools for data mining and analysis, such as classification, regression, clustering, dimensionality reduction, model selection and pre-processing;
-
LIBSVM82: an integrated software for support vector classification, regression, distribution estimation and multi-class classification;
-
LIBLINEAR83: a linear classifier for data with millions of instances and features;
-
SVM-Light84: is an implementation of SVMs for pattern recognition, classification, regression and ranking problems.
3.6.3 Opinion mining
-
SentiStrength85: a sentiment analysis tool that is able to conduct binary (positive/negative), trinary (positive/neutral/negative), single-scale (-4 very negative to very positive +4), keyword-oriented and domain-oriented classifications;
-
Sentiment14086: a tool that allows you to discover the sentiment of a brand, product, or topic on Twitter;
-
VADER (Valence Aware Dictionary and sEntiment Reasoner)87: a lexicon and rule-based sentiment analysis tool that is specifically focused on sentiments expressed in social media.
3.6.4 Big data
3.7 Natural language processing tasks
3.7.1 Overview
3.7.2 Pre-processing and negations
3.7.3 Emoticons/Emojis
3.7.4 Word embeddings
3.7.5 Aspect-based social opinion mining
4 Dimensions of social opinion mining
4.1 Context
4.2 Different dimensions of social opinions identified in the review analysis
Dimensions | Studies |
---|---|
subjectivity and sentiment polarity |
Jiang et al. (2011), Blenn et al. (2012), Bravo-Marquez et al. (2013), Zhu et al. (2013), Wang and Ye (2013), Cui et al. (2013), Li and Li (2013), Rui et al. (2013), Bravo-Marquez et al. (2014), Tan et al. (2014), Garg and Chatterjee (2014), Abdul-Mageed et al. (2014), Samoylov (2014), Koto and Adriani (2015), Koto and Adriani (2015), Koto and Adriani (2015), Feng et al. (2015), Mansour et al. (2015), Wu et al. (2016), Zainuddin et al. (2016), Er et al. (2016), Abdullah and Zolkepli (2017), Hao et al. (2017), Ahuja and Dubey (2017), Sahni et al. (2017), Moh et al. (2017), Dritsas et al. (2018), Gandhe et al. (2018) and Nausheen and Begum (2018) |
sentiment polarity and emotion |
Cvijikj and Michahelles (2011), Orellana-Rodriguez et al. (2013), Sheth et al. (2014), Yuan et al. (2015), Orellana-Rodriguez et al. (2015), Gallegos et al. (2016), Qaisi and Aljarah (2016), Shukri et al. (2015), Munezero et al. (2015), Barapatre et al. (2016), Karyotis et al. (2017), Bouazizi and Ohtsuki (2017), Radhika and Sankar (2017), Abdullah and Hadzikadic (2017), Zhang et al. (2017), Singh et al. (2018), Aoudi and Malik (2018), Pai and Alathur (2018), Ghosal et al. (2018), Rout et al. (2018), dos Santos et al. (2018) and Stojanovski et al. (2018) |
sentiment polarity and mood |
Bollen et al. (2011) |
sentiment polarity and irony |
Reyes et al. (2013) |
sentiment polarity and sarcasm |
Unankard et al. (2014) |
sentiment polarity and affect |
Weichselbraun et al. (2017) |
emotion and anger | |
irony and sarcasm |
Fersini et al. (2015) |
subjectivity, sentiment polarity and emotion |
Jiang et al. (2015) |
subjectivity, sentiment polarity, emotion and irony |
Bosco et al. (2013) |
4.2.1 Subjectivity
4.2.2 Sentiment
4.2.3 Emotion
4.2.4 Affect
4.2.5 Irony
4.2.6 Sarcasm
4.2.7 Mood
4.2.8 Aggressiveness
4.2.9 Other
4.3 Impact of sarcasm and irony on social opinions
4.3.1 Challenges
-
Different languages and cultures result in various ways of how an opinion is expressed on certain social media platforms. For example, Sina Weibo users prefer to use irony when expressing negative polarity (Wang et al. 2014). Future research is required for the development of cross-lingual/multilingual NLP tools that are able to identify irony and sarcasm (Yan et al. 2014).
-
Presence of sarcasm and irony in social data, such as tweets, may affect the feature values of certain machine learning algorithms. Therefore, further advancement is required in the techniques used for handling sarcastic and ironic tweets (Pandey et al. 2017). The work in Sarsam et al. (2020) addresses the main challenges faced for sarcasm detection in Twitter and the machine learning algorithms that can be used in this regard.
-
Classifying/rating a given sentence’s sentiment is very difficult and ambiguous, since people often use negative words to be humorous or sarcastic.
-
Sarcasm and/or irony annotation is very hard for humans and thus it should be presented to multiple persons for accuracy purposes. This makes it very challenging to collect large datasets that can be used for supervised learning, with the only possible way being to hire people to carry out such annotations (D’Asaro et al. 2017). Moreover, the differentiation between sarcasm and irony by human annotators result in a lack of available datasets and datasets with enough examples of ironic and/or sarcastic annotations. Such datasets are usually needed for “data hungry” computational learning methods (Sykora et al. 2020).
4.3.2 Observations
Sarcasm | Irony | Studies |
---|---|---|
✓ |
Baccouche et al. (2018), Bouazizi and Ohtsuki (2018), Ghiassi and Lee (2018), Abdullah and Zolkepli (2017), Bouazizi and Ohtsuki (2017), Caschera et al. (2016), Tan et al. (2014), Unankard et al. (2014), Mejova et al. (2013), Bakliwal et al. (2013), Mejova and Srinivasan (2012) and Wang et al. (2012) | |
✓ | ||
✓ | ✓ |
-
Bosco et al. (2013): The authors found that irony is normally used together with a positive statement to express a negative statement, but seldomly the other way. Analysis shows that the Senti-TUT106 corpus can be representative for a wide range of irony in phenomena from bitter sarcasm to genteel irony.
-
Reyes et al. (2013): The study describes a number of textual features used to identify irony at a linguistic level. These are mostly applicable for short texts, such as tweets. The developed irony detection model is evaluated in terms of representativeness and relevance. Authors also mention that there are overlaps in occurrences of irony, satire, parody and sarcasm, with their main differentiators being tied to usage, tone and obviousness.
-
Mejova et al. (2013): A multi-stage data-driven political sentiment classifier is proposed in this study. The authors found out “that a humorous tweet is 76.7% likely to also be sarcastic”, whereas “sarcastic tweets are only 26.2% likely to be humorous”. Future work is required on the connection between sarcasm and humour.
-
Fersini et al. (2015): Addresses the automatic detection of sarcasm and irony by introducing an ensemble approach based on Bayesian Model Averaging, that takes into account several classifiers according to their reliabilities and their marginal probability predictions. Results show that not all the features are equally able to characterise sarcasm and irony, whereby sarcasm is better characterised by POS tags, and ironic statements by pragmatic particles (such as emoticons and emphatic/onomatopoeic expressions, which represent those linguistic elements typically used in social media to convey a particular message).
-
Jiang et al. (2015): The authors’ model classifies subjectivity, polarity and emotion in microblogs. Results show that emoticons are a pure carrier of sentiment, whereas sentiment words have more complex senses and contexts, such as negations and irony.
-
Wang et al. (2012): Post-facto analysis of user-generated content, such as tweets, show that political tweets tend to be quite sarcastic.
-
Ghiassi and Lee (2018): Certain keywords or hash-tagged words (e.g., “thanks”, “#smh”, “ #not”) that follow certain negative or positive sentiment markers in textual social data, usually indicate the presence of sarcasm.
5 Application areas of social opinion mining
-
Telecommunications (e.g., telephony, television) on particular service providers (Ghiassi and Lee 2018; Ranjan et al. 2018; Napitu et al. 2017; Fitri et al. 2018; Kumar and Bala 2016; Varshney and Gupta 2014; Wunnasri et al. 2013; Tan et al. 2011; Trung et al. 2013) or complaints (Souza et al. 2016);
6 Concluding remarks
6.1 Latest research of social opinion mining
6.2 Conclusion
-
Early-Stage Researchers who are interested in working within this evolving research field of study and/or are looking for an overview of this field;
-
Experienced Researchers already working in SOM who would like to progress further on the technical side of their work and/or looking for weaknesses in the the field of SOM;
-
Early-Stage and/or Experienced Researchers who are looking into applying SOM/their SOM work in a real-world application area.
-
Social Media Platforms: Most studies focus on data gathered from one social media platform, with Twitter being the most popular followed by Sina Weibo for Chinese targeted studies. It is encouraged to possibly explore multi-source information by using other platforms, thus use data from multiple data sources, subject to any existing API limitations111. This shall increase the variety and volume of data (two of the V’s of Big Data) used for evaluation purposes, thus ensuring that results provide more reflective picture of society in terms of opinions. The use of multiple data sources for studies focusing on the same real-world application areas are also beneficial for comparison purposes and identification of any potential common traits, patterns and/or results. Mining opinions from multiple sources of information also presents several advantages, such as higher authenticity, reduced ambiguity and greater availability (Balazs and Velásquez 2016).
-
Techniques: The use of Deep Learning, Statistical, Probabilistic, Ontology and Graph-based approaches should be further explored both as standalone and/or part of hybrid techniques, due to their potential and accessibility. In particular, Deep Learning capabilities has made several applications feasible, whereas Ontologies and Graph Mining enable fine-grained opinion mining and the identification of relationships between opinions and their enablers (person, organisation, etc.). Moreover, ensemble Machine Learning and Deep Learning methods and fine-tuned Transformed-based models are still under-explored. In such a case, researchers should be attentive to the carbon footprint needed to train neural network models for NLP.
-
Social Datasets: The majority of available datasets are either English or Chinese specific. This domain needs further social datasets published under a common open license for use by the public domain. These should target any of the following criteria: bi-lingual/multilingual data, and/or annotations of multiple opinion dimensions within the data, e.g., sentiment polarity, emotion, sarcasm, irony, mood, etc. Both requirements are costly in terms of resources (time, funding and personnel), domain knowledge and expertise.
-
Language: The majority of the studies support one language, with English and Chinese being the most popular. Studies that support two or more languages is one of the major challenges in this domain due to numerous factors, such as cultural differences and lack of language-specific resources, e.g., lexicons, datasets, tools and technologies. This domain also needs more studies that focus on code-switched languages and less-resourced languages, which shall enable the development of certain language resources needed for the respective code-switched and less-resourced languages.
-
Modality: Bi-/Multi-modal SOM is another sub-domain that requires several research. Several studies cater for the text modality only, with the visual—image modality gaining more popularity. However, the visual—video and audio modalities are still in their early research phases with several aspects still unexplored. This also stems from a lack of available visual, audio and multimodal datasets.
-
Aspect-based SOM: Research in this sub-domain is increasing and developing, however, it is far from the finished article, especially when applied in certain domains. Further aspect-based research is encouraged on other opinion dimensions other than sentiment polarity, such as emotions and moods, which is still unexplored. Moreover, more research is required on the use of Deep Learning approaches for such a task, which is still at an early stage.
-
Application areas: Most studies target Politics, Marketing & Advertising & Sales, Technology, Finance, Film and Healthcare. Research into other areas/sub-domains is encouraged to study and show the potential of SOM.
-
Dimensions of SOM: Most studies focus on subjectivity detection and sentiment analysis. The area of emotion analysis is increasing in popularity, however, sarcasm detection, irony detection and mood analysis are still in their early research phases. Moreover, from the analysis of this systematic review it is evident that there is a lack of research on any possible correlations between the different opinion dimensions, e.g., emotions and sentiment. Lastly, no studies cater for all the different SOM dimensions within their work.