{ "paper_id": "2020", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T04:33:38.182377Z" }, "title": "A Dataset for Troll Classification of TamilMemes", "authors": [ { "first": "Shardul", "middle": [], "last": "Suryawanshi", "suffix": "", "affiliation": {}, "email": "shardul.suryawanshi@insight-centre.org" }, { "first": "Bharathi", "middle": [ "Raja" ], "last": "Chakravarthi", "suffix": "", "affiliation": {}, "email": "bharathi.raja@insight-centre.org" }, { "first": "", "middle": [], "last": "Pranav Varma", "suffix": "", "affiliation": { "laboratory": "", "institution": "National University of Ireland", "location": { "settlement": "Galway" } }, "email": "" }, { "first": "Mihael", "middle": [], "last": "Arcan", "suffix": "", "affiliation": {}, "email": "" }, { "first": "John", "middle": [ "P" ], "last": "Mccrae", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Paul", "middle": [], "last": "Buitelaar", "suffix": "", "affiliation": {}, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "Social media are interactive platforms that facilitate the creation or sharing of information, ideas or other forms of expression among people. This exchange is not free from offensive, trolling or malicious contents targeting users or communities. One way of trolling is by making memes, which in most cases combines an image with a concept or catchphrase. The challenge of dealing with memes is that they are region-specific and their meaning is often obscured in humour or sarcasm. To facilitate the computational modelling of trolling in the memes for Indian languages, we created a meme dataset for Tamil (TamilMemes). We annotated and released the dataset containing suspected trolls and not-troll memes. In this paper, we use the a image classification to address the difficulties involved in the classification of troll memes with the existing methods. We found that the identification of a troll meme with such an image classifier is not feasible which has been corroborated with precision, recall and F1-score.", "pdf_parse": { "paper_id": "2020", "_pdf_hash": "", "abstract": [ { "text": "Social media are interactive platforms that facilitate the creation or sharing of information, ideas or other forms of expression among people. This exchange is not free from offensive, trolling or malicious contents targeting users or communities. One way of trolling is by making memes, which in most cases combines an image with a concept or catchphrase. The challenge of dealing with memes is that they are region-specific and their meaning is often obscured in humour or sarcasm. To facilitate the computational modelling of trolling in the memes for Indian languages, we created a meme dataset for Tamil (TamilMemes). We annotated and released the dataset containing suspected trolls and not-troll memes. In this paper, we use the a image classification to address the difficulties involved in the classification of troll memes with the existing methods. We found that the identification of a troll meme with such an image classifier is not feasible which has been corroborated with precision, recall and F1-score.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "Traditional media content distribution channels such as television, radio or newspapers are monitored and scrutinized for their content. Nevertheless, social media platforms on the Internet opened the door for people to contribute, leave a comment on existing content without any moderation. Although most of the time, the internet users are harmless, some produce offensive content due to anonymity and freedom provided by social networks. Due to this freedom, people are becoming creative in their jokes by making memes. Although memes are meant to be humorous, sometimes it becomes threatening and offensive to specific people or community. On the Internet, a troll is a person who upsets or starts a hatred towards people or community. Trolling is the activity of posting a message via social media that is intended to be offensive, provocative, or menacing to distract which often has a digressive or off-topic content with the intent of provoking the audience (Bishop, 2013; Bishop, 2014 ; Mojica de la Vega and Ng, 2018; Suryawanshi et al., 2020) . Despite this growing body of research in natural language processing, identifying trolling in memes has yet to be investigated. One way to understand how meme varies from other image posts was studied by Wang and Wen (2015) . According to the authors, memes combine two images or are a combination of an image and a witty, catchy or sarcastic text. In this work, we treat this task as an image classification problem. Due to the large population in India, the issue has emerged in the context of recent events. There have been several threats towards people or communities from memes. This is a serious threat which shames people or spreads hatred towards people or a particular community (Kumar et al., 2018; Rani et al., 2020; Suryawanshi et al., 2020) . There have been several studies on moderating trolling, however, for a social media administrator memes are hard to monitor as they are region-specific. Furthermore, their meaning is often obscure due to fused image-text representation. The content in Indian memes might be written in English, in a native language (native or foreign script), or in a mixture of languages and scripts (Ranjan et al., 2016; Chakravarthi et al., 2018; Jose et al., 2020; Priyadharshini et al., 2020; Chakravarthi et al., 2020a; Chakravarthi et al., 2020b) . This adds another challenge to the meme classification problem. In Figure 1 , Example 1 is written in Tamil with two images and Example 2 is written in English and Tamil (Roman Script) with two images. In the first example, the meme is trolling about the \"Vim dis-washer\" soap. The information in Example 1 can be translated into English as \"the price of a lemon is five Rupees\", whereby the image below shows a crying person. Just after the crying person the text says \"The price of a Vim bar with the power of 100 Lemon is just 10 Rupees\". This is an example of opinion manipulation with trolling as it influences the user opinion about products, companies and politics. This kind of memes might be effective in two ways. On the one hand, it is easy for companies and political parties to gain popularity. On the other hand, the trolls can damage the reputation of the company name or political party name. Example 2 shows a funny meme; it shows that a guy is talking to a poor lady while the girl in the car is looking at them. The image below includes a popular Tamil comedy actor with a short text written beneath \"We also talk nicely to ladies to get into a relationship\". Even though there is a widespread culture of memes on the Internet, the research on the classification of memes is not studied well. There are no systematic studies on classifying memes in a troll or not-troll category. In this work, we describe a dataset for classifying memes in such categories. To do this, we have collected a set of original memes from volunteers. We present baseline results using convolutional neural network (CNN) approaches for image classification. We report our findings in precision, recall and F-score and publish the code for this work at https: //github.com/sharduls007/TamilMemes.", "cite_spans": [ { "start": 966, "end": 980, "text": "(Bishop, 2013;", "ref_id": "BIBREF1" }, { "start": 981, "end": 993, "text": "Bishop, 2014", "ref_id": "BIBREF2" }, { "start": 1009, "end": 1027, "text": "Vega and Ng, 2018;", "ref_id": "BIBREF29" }, { "start": 1028, "end": 1053, "text": "Suryawanshi et al., 2020)", "ref_id": "BIBREF37" }, { "start": 1260, "end": 1279, "text": "Wang and Wen (2015)", "ref_id": "BIBREF38" }, { "start": 1745, "end": 1765, "text": "(Kumar et al., 2018;", "ref_id": "BIBREF21" }, { "start": 1766, "end": 1784, "text": "Rani et al., 2020;", "ref_id": "BIBREF33" }, { "start": 1785, "end": 1810, "text": "Suryawanshi et al., 2020)", "ref_id": "BIBREF37" }, { "start": 2197, "end": 2218, "text": "(Ranjan et al., 2016;", "ref_id": "BIBREF34" }, { "start": 2219, "end": 2245, "text": "Chakravarthi et al., 2018;", "ref_id": "BIBREF4" }, { "start": 2246, "end": 2264, "text": "Jose et al., 2020;", "ref_id": "BIBREF20" }, { "start": 2265, "end": 2293, "text": "Priyadharshini et al., 2020;", "ref_id": "BIBREF32" }, { "start": 2294, "end": 2321, "text": "Chakravarthi et al., 2020a;", "ref_id": "BIBREF9" }, { "start": 2322, "end": 2349, "text": "Chakravarthi et al., 2020b)", "ref_id": "BIBREF10" } ], "ref_spans": [ { "start": 2419, "end": 2427, "text": "Figure 1", "ref_id": "FIGREF0" } ], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "A troll meme is an implicit image that intents to demean or offend an individual on the Internet. Based on the definition \"Trolling is the activity of posting a message via social media that tend to be offensive, provocative, or menacing (Bishop, 2013; Bishop, 2014 ; Mojica de la Vega and Ng, 2018)\". Their main function is to distract the audience with the intent of provoking them. We define troll memes as a meme, which contains offensive text and non-offensive images, offensive images with non-offensive text, sarcastically offensive text with non-offensive images, or sarcastic images with offensive text to provoke, distract, and has a digressive or off-topic content with intend to demean or offend particular people, group or race. Figure 2 shows examples of trolling memes, Example 3 is trolling the potato chip brand called Lays. The translation of the text is \"If you buy one packet of air, then 5 chips free\", with its intention to damage the company's reputation. Figure 2 illustrates examples of not-troll memes. The translation of Example 4 would be \"Sorry my friend (girl)\". As this example does not contain any provoking or offensive content and is even funny, it should be listed in the not-troll category. As a troll meme is directed towards someone, it is easy to find such content in the comments section or group chat of social media. For our work, we collected memes from volunteers who sent them through WhatsApp, a social media for chatting and creating a group chat. The suspected troll memes then have been verified and annotated manually by the annotators. As the users who sent these troll memes belong to the Tamil speaking population, all the troll memes are in Tamil. The general format of the meme is the image and Tamil text embedded within the image. Most of the troll memes comes from the state of Tamil Nadu, in India. The Tamil language, which has 75 million speakers, 1 belongs to the Dravidian language family (Rao and Lalitha Devi, 2013; Chakravarthi et al., 2019a; Chakravarthi et al., 2019b; Chakravarthi et al., 2019c) and is one of the 22 scheduled languages of India (Dash et al., 2015) . As these troll memes can have a negative psychological effect on an individual, a constraint has to be in place for such a conversation. In this work, we are attempting to identify such troll memes by providing a dataset and image classifier to identify these memes.", "cite_spans": [ { "start": 238, "end": 252, "text": "(Bishop, 2013;", "ref_id": "BIBREF1" }, { "start": 253, "end": 265, "text": "Bishop, 2014", "ref_id": "BIBREF2" }, { "start": 1952, "end": 1980, "text": "(Rao and Lalitha Devi, 2013;", "ref_id": "BIBREF35" }, { "start": 1981, "end": 2008, "text": "Chakravarthi et al., 2019a;", "ref_id": "BIBREF6" }, { "start": 2009, "end": 2036, "text": "Chakravarthi et al., 2019b;", "ref_id": "BIBREF7" }, { "start": 2037, "end": 2064, "text": "Chakravarthi et al., 2019c)", "ref_id": "BIBREF8" }, { "start": 2115, "end": 2134, "text": "(Dash et al., 2015)", "ref_id": "BIBREF13" } ], "ref_spans": [ { "start": 742, "end": 750, "text": "Figure 2", "ref_id": "FIGREF1" }, { "start": 979, "end": 987, "text": "Figure 2", "ref_id": "FIGREF1" } ], "eq_spans": [], "section": "Troll Meme", "sec_num": "2." }, { "text": "Trolling in social media for text has been studied extensively (Bishop, 2013; Bishop, 2014 ; Mojica de la Vega and Ng, 2018; Malmasi and Zampieri, 2017; Kumar et al., 2018; Kumar, 2019) . Opinion manipulation trolling (Mihaylov et al., 2015b; Mihaylov et al., 2015a) , troll comments in News Community (Mihaylov and Nakov, 2016) , and the role of political trolls (Atanasov et al., 2019) have been studied. All these considered the trolling on text-only media. However, meme consist of images or images with text.", "cite_spans": [ { "start": 63, "end": 77, "text": "(Bishop, 2013;", "ref_id": "BIBREF1" }, { "start": 78, "end": 90, "text": "Bishop, 2014", "ref_id": "BIBREF2" }, { "start": 125, "end": 152, "text": "Malmasi and Zampieri, 2017;", "ref_id": "BIBREF24" }, { "start": 153, "end": 172, "text": "Kumar et al., 2018;", "ref_id": "BIBREF21" }, { "start": 173, "end": 185, "text": "Kumar, 2019)", "ref_id": "BIBREF22" }, { "start": 218, "end": 242, "text": "(Mihaylov et al., 2015b;", "ref_id": "BIBREF28" }, { "start": 243, "end": 266, "text": "Mihaylov et al., 2015a)", "ref_id": "BIBREF27" }, { "start": 302, "end": 328, "text": "(Mihaylov and Nakov, 2016)", "ref_id": "BIBREF26" }, { "start": 364, "end": 387, "text": "(Atanasov et al., 2019)", "ref_id": "BIBREF0" } ], "ref_spans": [], "eq_spans": [], "section": "Related Work", "sec_num": "3." }, { "text": "A related research area is on offensive content detection. Various works in the recent years have investigated Offensive and Aggression content in text (Clarke and Grieve, 2017; Mathur et al., 2018; Nogueira dos Santos et al., 2018; Galery et al., 2018) . For images, Gandhi et al. (2019) deals with offensive images and non-compliant logos. They have developed a computer-vision driven offensive and non-compliant image detection algorithm that identifies the offensive content in the image. They have categorized images as offensive if it has nudity, sexually explicit content, abusive text, objects used to promote violence or racially inappropriate content. The classifier takes advantage of a pre-trained object detector to identify the type of object in the image and then sends the image to the unit which specializes in detecting objects in the image. The majority of memes do not contain nudity or explicit sexual content due to the moderation of social media on nudity. Hence unlike their research, we are trying to identify troll memes by using image features derived by use of a convolutional neural network. Hate speech is a subset of offensive language and datasets associated with hate speech have been collected from social media such as Twitter (Xiang et al., 2012) , Instagram (Hosseinmardi et al., 2015) , Yahoo (Nobata et al., 2016) , YouTube (Dinakar et al., 2012) . In all of these works, only text corpora have been used to detect trolling, offensive, aggression and hate speech. Nevertheless, for memes, there is no such dataset. For Indian language memes, it is not available as to our knowledge. We are the first to develop a meme dataset for Tamil, with troll or not-troll annotation.", "cite_spans": [ { "start": 152, "end": 177, "text": "(Clarke and Grieve, 2017;", "ref_id": "BIBREF11" }, { "start": 178, "end": 198, "text": "Mathur et al., 2018;", "ref_id": "BIBREF25" }, { "start": 199, "end": 232, "text": "Nogueira dos Santos et al., 2018;", "ref_id": "BIBREF31" }, { "start": 233, "end": 253, "text": "Galery et al., 2018)", "ref_id": "BIBREF15" }, { "start": 268, "end": 288, "text": "Gandhi et al. (2019)", "ref_id": "BIBREF16" }, { "start": 1262, "end": 1282, "text": "(Xiang et al., 2012)", "ref_id": "BIBREF39" }, { "start": 1295, "end": 1322, "text": "(Hosseinmardi et al., 2015)", "ref_id": "BIBREF18" }, { "start": 1331, "end": 1352, "text": "(Nobata et al., 2016)", "ref_id": "BIBREF30" }, { "start": 1363, "end": 1385, "text": "(Dinakar et al., 2012)", "ref_id": "BIBREF14" } ], "ref_spans": [], "eq_spans": [], "section": "Related Work", "sec_num": "3." }, { "text": "For our study, people provided memes voluntarily for our research. Additionally, all personal identifiable information such as usernames are deleted from this dataset. The annotators were warned about the trolling content before viewing the meme, and our instructions informed them that they could quit the annotation campaign anytime if they felt uncomfortable.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Ethics", "sec_num": "4.1." }, { "text": "To retrieve high-quality meme data that would likely to include trolling, we asked the volunteers to provide us with memes that they get in their social media platforms, like WhatsApp, Facebook, Instagram, and Pinterest. The data was collected between November 1, 2019, until January 15, 2019, from sixteen volunteers. We are not disclosing any personal information of the volunteers such as gender as per their will. Figure 3 shows an example of the collected memes. We removed duplicate memes, however, we kept memes that uses the same image but different text. This was a challenging task since the same meme could have different file names. Hence the same meme could be annotated by different annotators. Due to this, we checked manually and removed such duplicates before sending them to annotators. An example is shown in Figure 3 , where the same image with different text is used. Example 5 describes the image as \"can not understand what you are saying\", whereby Example 6 describes image as \"I am confused\". ", "cite_spans": [], "ref_spans": [ { "start": 418, "end": 426, "text": "Figure 3", "ref_id": "FIGREF2" }, { "start": 828, "end": 836, "text": "Figure 3", "ref_id": "FIGREF2" } ], "eq_spans": [], "section": "Data collection", "sec_num": "4.2." }, { "text": "After we obtained the memes, we presented this data to the annotators using Google Forms. To not over-burden the annotators, we provided ten memes per page and hundred memes per form. For each form, the annotators are asked to decide if a given meme is of category troll or not-troll. As a part of annotation guidelines, we gave multiple examples of troll memes and not-troll memes to the annotators. The annotation for these examples has been done by the an annotator who is considered as a expert as well as a native Tamil speaker. Each meme is assigned to two different annotators, a male and a female annotator. To ensure the quality of the annotations and due to the region-specific nature of the annotation task, only native speakers from Tamil Nadu, India were recruited as annotators. Although we are not disclosing the gender demographics of volunteers who provided memes, we have gender-balanced annotation since each meme has been annotated by a male and a female.A meme is considered as troll only when both of the annotators label it as a troll.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Annotation", "sec_num": "4.3." }, { "text": "In order to evaluate the reliability of the annotation and their robustness across experiments, we analyzed the interannotator agreement using Cohen's kappa (Cohen, 1960) . It compares the probability of two annotators agreeing by chance with the observed agreement. It measures agreement expected by chance by modelling each annotator with separate distribution governing their likelihood of assigning a particular category. Mathematically,", "cite_spans": [ { "start": 157, "end": 170, "text": "(Cohen, 1960)", "ref_id": "BIBREF12" } ], "ref_spans": [], "eq_spans": [], "section": "Inter-Annotator Agreement", "sec_num": "4.4." }, { "text": "K = p(A) \u2212 p(E) 1 \u2212 p(E) (1)", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Inter-Annotator Agreement", "sec_num": "4.4." }, { "text": "where K is the kappa value, p(A) is the probability of the actual outcome and p(E) is the probability of the expected outcome as predicted by chance (Bloodgood and Grothendieck, 2013) . We got a kappa value of 0.62 between two annotators (gender balance male and female annotators). Based on Landis and Koch (1977) and given the inherent obscure nature of memes, we got fair agreement amongst the annotators.", "cite_spans": [ { "start": 149, "end": 183, "text": "(Bloodgood and Grothendieck, 2013)", "ref_id": "BIBREF3" }, { "start": 292, "end": 314, "text": "Landis and Koch (1977)", "ref_id": "BIBREF23" } ], "ref_spans": [], "eq_spans": [], "section": "Inter-Annotator Agreement", "sec_num": "4.4." }, { "text": "We collected 2,969 memes, of which most are images with text embedded on them. After the annotation, we learned that the majority (1,951) of these were annotated as troll memes, and 1,018 as not-troll memes. Furthermore, we observed that memes, which have more than one image have a high probability of being a troll, whereas those with only one image are likely to be not-troll. We included Flickr30K 2 images (Young et al., 2014) to the not-troll category to address the class imbalance. Flickr30K is only added to training, while the test set is randomly chosen from our dataset. In all our experiments the test set remains the same.", "cite_spans": [ { "start": 411, "end": 431, "text": "(Young et al., 2014)", "ref_id": "BIBREF40" } ], "ref_spans": [], "eq_spans": [], "section": "Data Statistics", "sec_num": "4.5." }, { "text": "To demonstrate how the given dataset can be used to classify troll memes, we defined two experiments with four variations of each. We measured the performance of the proposed baselines by using precision, recall and F1-score for each class, i.e. \"troll and not-troll\". We used ResNet (He et al., 2016) and MobileNet (Howard et al., 2017) as a baseline to perform the experiments. We give insights into their architecture and design choices in the sections below.", "cite_spans": [ { "start": 284, "end": 301, "text": "(He et al., 2016)", "ref_id": "BIBREF17" }, { "start": 316, "end": 337, "text": "(Howard et al., 2017)", "ref_id": "BIBREF19" } ], "ref_spans": [], "eq_spans": [], "section": "Methodology", "sec_num": "5." }, { "text": "ResNet has won the ImageNet ILSVRC 2015 (Russakovsky et al., 2015) classification task. It is still a popular method for classifying images and uses residual learning which connects low-level and high-level representation directly by skipping the connections in-between. This improves the performance of ResNet by diminishing the problem of vanishing gradient descent. It assumes that a deeper network should not produce higher training error than a shallow network. In this experiment, we used the ResNet architecture with 176 layers. As it was trained on the ImageNet task, we removed the classification (last) layer and used GlobalAveragePooling in place of fully connected layer to save the computational cost. Later, we added four fully connected layers with the classification layer which has a sigmoid activation function.This architecture is trained with or without pre-trained ImageNet weights.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "ResNet", "sec_num": null }, { "text": "We trained MobileNet with and without ImageNet weights. The model has a depth multiplier of 1.4, and an input dimension of 224\u00d7224 pixels. This provides a 1, 280\u00d71.4 = 1, 792 -dimensional representation of an image, which is then passed through a single hidden layer of a dimensionality of 1, 024 with ReLU activation, before being passed to a hidden layer with input dimension of (512,None) without any activation to provide the final representation h p . The main purpose of MobileNet is to optimize convolutional neural networks for mobile and embedded vision applications. It is less complex than ResNet in terms of number of hyperparameters and operations. It uses a different convolutional layer for each channel, this allows parallel computation on each channel which is Depthwise Separable Convolution. Later on the features extracted from these layers have been combined using the pointwise convolution layer. We used MobileNet to reduce the computational cost and compare it with the computationally intensive ResNet.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "MobileNet", "sec_num": null }, { "text": "We experimented with ResNet and MobileNet. The variation in experiments comes in terms of the data on which the models have been trained on, while the test set (300 memes) remained the same for all experiments. In the first variation, TamilMemes in Table 1 , we trained the ResNet and Mo-bileNet models on our Tamil meme dataset(2,669 memes).", "cite_spans": [], "ref_spans": [ { "start": 249, "end": 256, "text": "Table 1", "ref_id": "TABREF0" } ], "eq_spans": [], "section": "Experiments", "sec_num": "6." }, { "text": "The second variation, i.e. TamilMemes + ImageNet uses pre-trained ImageNet weights on the Tamil memes dataset.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Experiments", "sec_num": "6." }, { "text": "To address the class imbalance, we added 1,000 images from the Flickr30k dataset to the training set in the third variation i.e. TamilMemes + ImageNet + Flickr1k. As a result, the third variation has 3,969 images (1,951 trolls and 2,018 not-trolls). In the last variation, TamilMemes + ImageNet + Flickr30k, we added 30,000 images from the Flickr30k dataset to not-troll category. Flickr dataset has images and the captions which describes the image. We used these images as a not-troll category because they do not convey trollings without the context of the text. Except for the TamilMemes baseline, we are using pre-trained Im-ageNet weights for all other variations. Images from the Flickr30k dataset are used to balance the not-troll class in the TamilMemes + ImageNet + Flickr1k variation. On the one hand, the use of all the samples from the Flickr30k dataset as not-troll in the fourth variation introduces the class imbalance by significantly increasing the number of not-troll samples compared to the troll one. On the other hand, in the first variation, a higher number of troll meme samples again introduces a class imbalance.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Experiments", "sec_num": "6." }, { "text": "In the ResNet variations, we observed that there is no change in the macro averaged precision, recall and F1-score except for TamilMemes + ImageNet + Flickr1k variation. This variation has relatively poor results when compared with the other three variations in ResNet. While precision at identifying the troll class for the ResNet baseline does not vary much, we get better precision at classifying troll memes in the TamilMemes variation. This shows that the ResNet model trained on just Tamil memes has a better chance at identifying troll memes. The scenario is different in the case of the MobileNet variations. On the one hand, we observed less precision at identifying the troll class for the TamilMemes variation. On the other hand, we see improvement in precision at detecting trolls in the TamilMeme + ImageNet variation. This shows that MobileNet can leverage transfer learning to improve results. The relatively poor performance of MobileNet on the TamilMeme variation shows that it can not learn complex features like ResNet does to identify troll memes. For ResNet, the trend in the macro averaged score can be seen increasing in TamilMemes + ImageNet and TamilMemes + ImageNet + Flickr1k variations when compared to the TamilMemes variation. The TamilMemes + ImageNet + Flickr30k variation shows a lower macro averaged score than that of the TamilMemes + ImageNet + Flickr1k variation in both MobileNet and ResNet. Overall the precision for troll class identification lies in the range of 0.28 and 0.37, which is rather less than that of the not-troll class which lies in the range of 0.64 and 0.68. When we train ResNet in class imbalanced data in TamilMemes and TamilMemes + ImageNet + Flickr30k variations, results shows that the macro-averaged score of these variations are not hampered by the class imbalance issue. While for same variations MobileNet shows poor macro-averaged precision and recall score when compared with other variations. This shows that MobileNet is more susceptible to class imbalance issue than ResNet.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Result and Discussion", "sec_num": "7." }, { "text": "As shown in the Table 1 the classification model performs poorly at identifying of troll memes. We observed that this stems from the problem characteristics of memes. The meme dataset is unbalanced and memes have both image and text embedded to it with code-mixing in different forms. Therefore, it is inherently more challenging to train a classifier using just images. Further, the same image can be used with different text to mean different things, potentially making the task more complicated.", "cite_spans": [], "ref_spans": [ { "start": 16, "end": 23, "text": "Table 1", "ref_id": "TABREF0" } ], "eq_spans": [], "section": "Conclusions and Future work", "sec_num": "8." }, { "text": "To reduce the burden placed on annotators, we plan to use a semi-supervised approach to the size of the dataset. Semisupervised approaches have been proven to be of good use to increase the size of the datasets for under-resourced scenarios. We plan to use optical character recognizer (OCR) followed by a manual evaluation to obtain the text in the images. Since Tamil memes have code-mixing phenomenon, we plan to tackle the problem accordingly. With text identification using OCR, we will be able to approach the problem in a multi-modal way. We have created a meme dataset only for Tamil, but we plan to extend this to other languages as well.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusions and Future work", "sec_num": "8." }, { "text": "https://www.ethnologue.com/language/tam", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "https://github.com/BryanPlummer/flickr30k entities", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [ { "text": "This publication has emanated from research supported in part by a research grant from Science Foundation Ireland (SFI) under Grant Number SFI/12/RC/2289 P2, co-funded by the European Regional Development Fund, as well as by the H2020 project Pr\u00eat-\u00e0-LLOD under Grant Agreement number 825182.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Acknowledgments", "sec_num": null } ], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "Predicting the role of political trolls in social media", "authors": [ { "first": "A", "middle": [], "last": "Atanasov", "suffix": "" }, { "first": "G", "middle": [], "last": "De Francisci Morales", "suffix": "" }, { "first": "P", "middle": [], "last": "Nakov", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL)", "volume": "", "issue": "", "pages": "1023--1034", "other_ids": {}, "num": null, "urls": [], "raw_text": "Atanasov, A., De Francisci Morales, G., and Nakov, P. (2019). Predicting the role of political trolls in social media. In Proceedings of the 23rd Conference on Com- putational Natural Language Learning (CoNLL), pages 1023-1034, Hong Kong, China, November. Association for Computational Linguistics.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "The effect of de-individuation of the internet troller on criminal procedure implementation: An interview with a hater -proquest", "authors": [ { "first": "J", "middle": [], "last": "Bishop", "suffix": "" } ], "year": 2013, "venue": "International journal of cyber criminology", "volume": "", "issue": "", "pages": "28--48", "other_ids": {}, "num": null, "urls": [], "raw_text": "Bishop, J. (2013). The effect of de-individuation of the in- ternet troller on criminal procedure implementation: An interview with a hater -proquest. International journal of cyber criminology, page 28-48.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Dealing with internet trolling in political online communities: Towards the this is why we can't have nice things scale", "authors": [ { "first": "J", "middle": [], "last": "Bishop", "suffix": "" } ], "year": 2014, "venue": "Int. J. E-Polit", "volume": "5", "issue": "4", "pages": "1--20", "other_ids": {}, "num": null, "urls": [], "raw_text": "Bishop, J. (2014). Dealing with internet trolling in politi- cal online communities: Towards the this is why we can't have nice things scale. Int. J. E-Polit., 5(4):1-20, Octo- ber.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Analysis of stopping active learning based on stabilizing predictions", "authors": [ { "first": "M", "middle": [], "last": "Bloodgood", "suffix": "" }, { "first": "J", "middle": [], "last": "Grothendieck", "suffix": "" } ], "year": 2013, "venue": "Proceedings of the Seventeenth Conference on Computational Natural Language Learning", "volume": "", "issue": "", "pages": "10--19", "other_ids": {}, "num": null, "urls": [], "raw_text": "Bloodgood, M. and Grothendieck, J. (2013). Analysis of stopping active learning based on stabilizing predictions. In Proceedings of the Seventeenth Conference on Com- putational Natural Language Learning, pages 10-19, Sofia, Bulgaria, August. Association for Computational Linguistics.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Improving Wordnets for Under-Resourced Languages Using Machine Translation", "authors": [ { "first": "B", "middle": [ "R" ], "last": "Chakravarthi", "suffix": "" }, { "first": "M", "middle": [], "last": "Arcan", "suffix": "" }, { "first": "J", "middle": [ "P" ], "last": "Mccrae", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the 9th", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Chakravarthi, B. R., Arcan, M., and McCrae, J. P. (2018). Improving Wordnets for Under-Resourced Languages Using Machine Translation. In Proceedings of the 9th", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Global WordNet Conference. The Global WordNet Conference", "authors": [], "year": 2018, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Global WordNet Conference. The Global WordNet Con- ference 2018 Committee.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Comparison of Different Orthographies for Machine Translation of Under-Resourced Dravidian Languages", "authors": [ { "first": "B", "middle": [ "R" ], "last": "Chakravarthi", "suffix": "" }, { "first": "M", "middle": [], "last": "Arcan", "suffix": "" }, { "first": "J", "middle": [ "P" ], "last": "Mccrae", "suffix": "" } ], "year": 2019, "venue": "2nd Conference on Language, Data and Knowledge (LDK 2019)", "volume": "6", "issue": "", "pages": "1--6", "other_ids": {}, "num": null, "urls": [], "raw_text": "Chakravarthi, B. R., Arcan, M., and McCrae, J. P. (2019a). Comparison of Different Orthographies for Machine Translation of Under-Resourced Dravidian Languages. In Maria Eskevich, et al., editors, 2nd Conference on Language, Data and Knowledge (LDK 2019), vol- ume 70 of OpenAccess Series in Informatics (OASIcs), pages 6:1-6:14, Dagstuhl, Germany. Schloss Dagstuhl- Leibniz-Zentrum fuer Informatik.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "WordNet gloss translation for under-resourced languages using multilingual neural machine translation", "authors": [ { "first": "B", "middle": [ "R" ], "last": "Chakravarthi", "suffix": "" }, { "first": "M", "middle": [], "last": "Arcan", "suffix": "" }, { "first": "J", "middle": [ "P" ], "last": "Mccrae", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the Second Workshop on Multilingualism at the Intersection of Knowledge Bases and Machine Translation", "volume": "", "issue": "", "pages": "1--7", "other_ids": {}, "num": null, "urls": [], "raw_text": "Chakravarthi, B. R., Arcan, M., and McCrae, J. P. (2019b). WordNet gloss translation for under-resourced languages using multilingual neural machine translation. In Pro- ceedings of the Second Workshop on Multilingualism at the Intersection of Knowledge Bases and Machine Trans- lation, pages 1-7, Dublin, Ireland, August. European As- sociation for Machine Translation.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "Multilingual multimodal machine translation for Dravidian languages utilizing phonetic transcription", "authors": [ { "first": "B", "middle": [ "R" ], "last": "Chakravarthi", "suffix": "" }, { "first": "R", "middle": [], "last": "Priyadharshini", "suffix": "" }, { "first": "B", "middle": [], "last": "Stearns", "suffix": "" }, { "first": "A", "middle": [], "last": "Jayapal", "suffix": "" }, { "first": "S", "middle": [], "last": "", "suffix": "" }, { "first": "S", "middle": [], "last": "Arcan", "suffix": "" }, { "first": "M", "middle": [], "last": "Zarrouk", "suffix": "" }, { "first": "M", "middle": [], "last": "Mccrae", "suffix": "" }, { "first": "J", "middle": [ "P" ], "last": "", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the 2nd Workshop on Technologies for MT of Low Resource Languages", "volume": "", "issue": "", "pages": "56--63", "other_ids": {}, "num": null, "urls": [], "raw_text": "Chakravarthi, B. R., Priyadharshini, R., Stearns, B., Jaya- pal, A., S, S., Arcan, M., Zarrouk, M., and McCrae, J. P. (2019c). Multilingual multimodal machine transla- tion for Dravidian languages utilizing phonetic transcrip- tion. In Proceedings of the 2nd Workshop on Technolo- gies for MT of Low Resource Languages, pages 56-63, Dublin, Ireland, August. European Association for Ma- chine Translation.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "A sentiment analysis dataset for code-mixed Malayalam-English", "authors": [ { "first": "B", "middle": [ "R" ], "last": "Chakravarthi", "suffix": "" }, { "first": "N", "middle": [], "last": "Jose", "suffix": "" }, { "first": "S", "middle": [], "last": "Suryawanshi", "suffix": "" }, { "first": "E", "middle": [], "last": "Sherly", "suffix": "" }, { "first": "J", "middle": [ "P" ], "last": "Mccrae", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the 1st Joint Workshop of SLTU (Spoken Language Technologies for Under-resourced languages) and CCURL (Collaboration and Computing for Under-Resourced Languages) (SLTU-CCURL 2020)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Chakravarthi, B. R., Jose, N., Suryawanshi, S., Sherly, E., and McCrae, J. P. (2020a). A sentiment analy- sis dataset for code-mixed Malayalam-English. In Pro- ceedings of the 1st Joint Workshop of SLTU (Spoken Language Technologies for Under-resourced languages) and CCURL (Collaboration and Computing for Under- Resourced Languages) (SLTU-CCURL 2020), Marseille, France, May. European Language Resources Association (ELRA).", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "Corpus creation for sentiment analysis in code-mixed Tamil-English text", "authors": [ { "first": "B", "middle": [ "R" ], "last": "Chakravarthi", "suffix": "" }, { "first": "V", "middle": [], "last": "Muralidaran", "suffix": "" }, { "first": "R", "middle": [], "last": "Priyadharshini", "suffix": "" }, { "first": "J", "middle": [ "P" ], "last": "Mccrae", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the 1st Joint Workshop of SLTU (Spoken Language Technologies for Under-resourced languages) and CCURL (Collaboration and Computing for Under-Resourced Languages) (SLTU-CCURL 2020)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Chakravarthi, B. R., Muralidaran, V., Priyadharshini, R., and McCrae, J. P. (2020b). Corpus creation for senti- ment analysis in code-mixed Tamil-English text. In Pro- ceedings of the 1st Joint Workshop of SLTU (Spoken Language Technologies for Under-resourced languages) and CCURL (Collaboration and Computing for Under- Resourced Languages) (SLTU-CCURL 2020), Marseille, France, May. European Language Resources Association (ELRA).", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "Dimensions of abusive language on Twitter", "authors": [ { "first": "I", "middle": [], "last": "Clarke", "suffix": "" }, { "first": "J", "middle": [], "last": "Grieve", "suffix": "" } ], "year": 2017, "venue": "Proceedings of the First Workshop on Abusive Language Online", "volume": "", "issue": "", "pages": "1--10", "other_ids": {}, "num": null, "urls": [], "raw_text": "Clarke, I. and Grieve, J. (2017). Dimensions of abusive language on Twitter. In Proceedings of the First Work- shop on Abusive Language Online, pages 1-10, Vancou- ver, BC, Canada, August. Association for Computational Linguistics.", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "A Coefficient of Agreement for Nominal Scales. Educational and Psychological Measurement", "authors": [ { "first": "J", "middle": [], "last": "Cohen", "suffix": "" } ], "year": 1960, "venue": "", "volume": "20", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Cohen, J. (1960). A Coefficient of Agreement for Nom- inal Scales. Educational and Psychological Measure- ment, 20(1):37.", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "Generating translation corpora in Indic languages:cultivating bilingual texts for cross lingual fertilization", "authors": [ { "first": "N", "middle": [ "S" ], "last": "Dash", "suffix": "" }, { "first": "A", "middle": [], "last": "Selvraj", "suffix": "" }, { "first": "M", "middle": [], "last": "Hussain", "suffix": "" } ], "year": 2015, "venue": "Proceedings of the 12th International Conference on Natural Language Processing", "volume": "", "issue": "", "pages": "333--342", "other_ids": {}, "num": null, "urls": [], "raw_text": "Dash, N. S., Selvraj, A., and Hussain, M. (2015). Gen- erating translation corpora in Indic languages:cultivating bilingual texts for cross lingual fertilization. In Proceed- ings of the 12th International Conference on Natural Language Processing, pages 333-342, Trivandrum, In- dia, December. NLP Association of India.", "links": null }, "BIBREF14": { "ref_id": "b14", "title": "Common sense reasoning for detection, prevention, and mitigation of cyberbullying", "authors": [ { "first": "K", "middle": [], "last": "Dinakar", "suffix": "" }, { "first": "B", "middle": [], "last": "Jones", "suffix": "" }, { "first": "C", "middle": [], "last": "Havasi", "suffix": "" }, { "first": "H", "middle": [], "last": "Lieberman", "suffix": "" }, { "first": "R", "middle": [], "last": "Picard", "suffix": "" } ], "year": 2012, "venue": "ACM Trans. Interact. Intell. Syst", "volume": "2", "issue": "3", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Dinakar, K., Jones, B., Havasi, C., Lieberman, H., and Pi- card, R. (2012). Common sense reasoning for detection, prevention, and mitigation of cyberbullying. ACM Trans. Interact. Intell. Syst., 2(3), September.", "links": null }, "BIBREF15": { "ref_id": "b15", "title": "Aggression identification and multi lingual word embeddings", "authors": [ { "first": "T", "middle": [], "last": "Galery", "suffix": "" }, { "first": "E", "middle": [], "last": "Charitos", "suffix": "" }, { "first": "Y", "middle": [], "last": "Tian", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the First Workshop on Trolling, Aggression and Cyberbullying (TRAC-2018)", "volume": "", "issue": "", "pages": "74--79", "other_ids": {}, "num": null, "urls": [], "raw_text": "Galery, T., Charitos, E., and Tian, Y. (2018). Aggres- sion identification and multi lingual word embeddings. In Proceedings of the First Workshop on Trolling, Ag- gression and Cyberbullying (TRAC-2018), pages 74-79, Santa Fe, New Mexico, USA, August. Association for Computational Linguistics.", "links": null }, "BIBREF16": { "ref_id": "b16", "title": "Image matters: Detecting offensive and non-compliant content/logo in product images", "authors": [ { "first": "S", "middle": [], "last": "Gandhi", "suffix": "" }, { "first": "S", "middle": [], "last": "Kokkula", "suffix": "" }, { "first": "A", "middle": [], "last": "Chaudhuri", "suffix": "" }, { "first": "A", "middle": [], "last": "Magnani", "suffix": "" }, { "first": "T", "middle": [], "last": "Stanley", "suffix": "" }, { "first": "B", "middle": [], "last": "Ahmadi", "suffix": "" }, { "first": "V", "middle": [], "last": "Kandaswamy", "suffix": "" }, { "first": "O", "middle": [], "last": "Ovenc", "suffix": "" }, { "first": "S", "middle": [], "last": "Mannor", "suffix": "" } ], "year": 2019, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": { "arXiv": [ "arXiv:1905.02234" ] }, "num": null, "urls": [], "raw_text": "Gandhi, S., Kokkula, S., Chaudhuri, A., Magnani, A., Stan- ley, T., Ahmadi, B., Kandaswamy, V., Ovenc, O., and Mannor, S. (2019). Image matters: Detecting offen- sive and non-compliant content/logo in product images. arXiv preprint arXiv:1905.02234.", "links": null }, "BIBREF17": { "ref_id": "b17", "title": "Deep residual learning for image recognition", "authors": [ { "first": "K", "middle": [], "last": "He", "suffix": "" }, { "first": "X", "middle": [], "last": "Zhang", "suffix": "" }, { "first": "S", "middle": [], "last": "Ren", "suffix": "" }, { "first": "J", "middle": [], "last": "Sun", "suffix": "" } ], "year": 2016, "venue": "2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016", "volume": "", "issue": "", "pages": "770--778", "other_ids": {}, "num": null, "urls": [], "raw_text": "He, K., Zhang, X., Ren, S., and Sun, J. (2016). Deep residual learning for image recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recogni- tion, CVPR 2016, Las Vegas, NV, USA, June 27-30, 2016, pages 770-778.", "links": null }, "BIBREF18": { "ref_id": "b18", "title": "Analyzing labeled cyberbullying incidents on the instagram social network", "authors": [ { "first": "H", "middle": [], "last": "Hosseinmardi", "suffix": "" }, { "first": "S", "middle": [ "A" ], "last": "Mattson", "suffix": "" }, { "first": "R", "middle": [ "I" ], "last": "Rafiq", "suffix": "" }, { "first": "R", "middle": [], "last": "Han", "suffix": "" }, { "first": "Q", "middle": [], "last": "Lv", "suffix": "" }, { "first": "S", "middle": [], "last": "Mishra", "suffix": "" } ], "year": 2015, "venue": "International conference on social informatics", "volume": "", "issue": "", "pages": "49--66", "other_ids": {}, "num": null, "urls": [], "raw_text": "Hosseinmardi, H., Mattson, S. A., Rafiq, R. I., Han, R., Lv, Q., and Mishra, S. (2015). Analyzing labeled cyberbul- lying incidents on the instagram social network. In Inter- national conference on social informatics, pages 49-66. Springer.", "links": null }, "BIBREF19": { "ref_id": "b19", "title": "Mobilenets: Efficient convolutional neural networks for mobile vision applications", "authors": [ { "first": "A", "middle": [ "G" ], "last": "Howard", "suffix": "" }, { "first": "M", "middle": [], "last": "Zhu", "suffix": "" }, { "first": "B", "middle": [], "last": "Chen", "suffix": "" }, { "first": "D", "middle": [], "last": "Kalenichenko", "suffix": "" }, { "first": "W", "middle": [], "last": "Wang", "suffix": "" }, { "first": "T", "middle": [], "last": "Weyand", "suffix": "" }, { "first": "M", "middle": [], "last": "Andreetto", "suffix": "" }, { "first": "Adam", "middle": [], "last": "", "suffix": "" }, { "first": "H", "middle": [], "last": "", "suffix": "" } ], "year": 2017, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": { "arXiv": [ "arXiv:1704.04861" ] }, "num": null, "urls": [], "raw_text": "Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., and Adam, H. (2017). Mobilenets: Efficient convolutional neural net- works for mobile vision applications. arXiv preprint arXiv:1704.04861.", "links": null }, "BIBREF20": { "ref_id": "b20", "title": "A survey of current datasets for code-switching research", "authors": [ { "first": "N", "middle": [], "last": "Jose", "suffix": "" }, { "first": "B", "middle": [ "R" ], "last": "Chakravarthi", "suffix": "" }, { "first": "S", "middle": [], "last": "Suryawanshi", "suffix": "" }, { "first": "E", "middle": [], "last": "Sherly", "suffix": "" }, { "first": "J", "middle": [ "P" ], "last": "Mccrae", "suffix": "" } ], "year": 2020, "venue": "2020 6th International Conference on Advanced Computing and Communication Systems (ICACCS)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Jose, N., Chakravarthi, B. R., Suryawanshi, S., Sherly, E., and McCrae, J. P. (2020). A survey of current datasets for code-switching research. In 2020 6th International Conference on Advanced Computing and Communica- tion Systems (ICACCS).", "links": null }, "BIBREF21": { "ref_id": "b21", "title": "Benchmarking aggression identification in social media", "authors": [ { "first": "R", "middle": [], "last": "Kumar", "suffix": "" }, { "first": "A", "middle": [ "K" ], "last": "Ojha", "suffix": "" }, { "first": "S", "middle": [], "last": "Malmasi", "suffix": "" }, { "first": "M", "middle": [], "last": "Zampieri", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the First Workshop on Trolling, Aggression and Cyberbullying (TRAC-2018)", "volume": "", "issue": "", "pages": "1--11", "other_ids": {}, "num": null, "urls": [], "raw_text": "Kumar, R., Ojha, A. K., Malmasi, S., and Zampieri, M. (2018). Benchmarking aggression identification in so- cial media. In Proceedings of the First Workshop on Trolling, Aggression and Cyberbullying (TRAC-2018), pages 1-11, Santa Fe, New Mexico, USA, August. As- sociation for Computational Linguistics.", "links": null }, "BIBREF22": { "ref_id": "b22", "title": "# shutdownjnu vs# standwithjnu: A study of aggression and conflict in political debates on social media in india", "authors": [ { "first": "R", "middle": [], "last": "Kumar", "suffix": "" } ], "year": 2019, "venue": "Journal of Language Aggression and Conflict", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Kumar, R. (2019). # shutdownjnu vs# standwithjnu: A study of aggression and conflict in political debates on social media in india. Journal of Language Aggression and Conflict.", "links": null }, "BIBREF23": { "ref_id": "b23", "title": "The measurement of observer agreement for categorical data", "authors": [ { "first": "J", "middle": [ "R" ], "last": "Landis", "suffix": "" }, { "first": "G", "middle": [ "G" ], "last": "Koch", "suffix": "" } ], "year": 1977, "venue": "Biometrics", "volume": "33", "issue": "1", "pages": "159--174", "other_ids": {}, "num": null, "urls": [], "raw_text": "Landis, J. R. and Koch, G. G. (1977). The measurement of observer agreement for categorical data. Biometrics, 33(1):159-174.", "links": null }, "BIBREF24": { "ref_id": "b24", "title": "Detecting hate speech in social media", "authors": [ { "first": "S", "middle": [], "last": "Malmasi", "suffix": "" }, { "first": "M", "middle": [], "last": "Zampieri", "suffix": "" } ], "year": 2017, "venue": "Proceedings of the International Conference Recent Advances in Natural Language Processing", "volume": "", "issue": "", "pages": "467--472", "other_ids": {}, "num": null, "urls": [], "raw_text": "Malmasi, S. and Zampieri, M. (2017). Detecting hate speech in social media. In Proceedings of the Interna- tional Conference Recent Advances in Natural Language Processing, RANLP 2017, pages 467-472, Varna, Bul- garia, September. INCOMA Ltd.", "links": null }, "BIBREF25": { "ref_id": "b25", "title": "Detecting offensive tweets in Hindi-English code-switched language", "authors": [ { "first": "P", "middle": [], "last": "Mathur", "suffix": "" }, { "first": "R", "middle": [], "last": "Shah", "suffix": "" }, { "first": "R", "middle": [], "last": "Sawhney", "suffix": "" }, { "first": "D", "middle": [], "last": "Mahata", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the Sixth International Workshop on Natural Language Processing for Social Media", "volume": "", "issue": "", "pages": "18--26", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mathur, P., Shah, R., Sawhney, R., and Mahata, D. (2018). Detecting offensive tweets in Hindi-English code-switched language. In Proceedings of the Sixth In- ternational Workshop on Natural Language Processing for Social Media, pages 18-26, Melbourne, Australia, July. Association for Computational Linguistics.", "links": null }, "BIBREF26": { "ref_id": "b26", "title": "Hunting for troll comments in news community forums", "authors": [ { "first": "T", "middle": [], "last": "Mihaylov", "suffix": "" }, { "first": "P", "middle": [], "last": "Nakov", "suffix": "" } ], "year": 2016, "venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics", "volume": "2", "issue": "", "pages": "399--405", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mihaylov, T. and Nakov, P. (2016). Hunting for troll com- ments in news community forums. In Proceedings of the 54th Annual Meeting of the Association for Computa- tional Linguistics (Volume 2: Short Papers), pages 399- 405, Berlin, Germany, August. Association for Compu- tational Linguistics.", "links": null }, "BIBREF27": { "ref_id": "b27", "title": "Finding opinion manipulation trolls in news community forums", "authors": [ { "first": "T", "middle": [], "last": "Mihaylov", "suffix": "" }, { "first": "G", "middle": [], "last": "Georgiev", "suffix": "" }, { "first": "P", "middle": [], "last": "Nakov", "suffix": "" } ], "year": 2015, "venue": "Proceedings of the Nineteenth Conference on Computational Natural Language Learning", "volume": "", "issue": "", "pages": "310--314", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mihaylov, T., Georgiev, G., and Nakov, P. (2015a). Find- ing opinion manipulation trolls in news community fo- rums. In Proceedings of the Nineteenth Conference on Computational Natural Language Learning, pages 310- 314, Beijing, China, July. Association for Computational Linguistics.", "links": null }, "BIBREF28": { "ref_id": "b28", "title": "Exposing paid opinion manipulation trolls", "authors": [ { "first": "T", "middle": [], "last": "Mihaylov", "suffix": "" }, { "first": "I", "middle": [], "last": "Koychev", "suffix": "" }, { "first": "G", "middle": [], "last": "Georgiev", "suffix": "" }, { "first": "P", "middle": [], "last": "Nakov", "suffix": "" } ], "year": 2015, "venue": "Proceedings of the International Conference Recent Advances in Natural Language Processing", "volume": "", "issue": "", "pages": "443--450", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mihaylov, T., Koychev, I., Georgiev, G., and Nakov, P. (2015b). Exposing paid opinion manipulation trolls. In Proceedings of the International Conference Recent Ad- vances in Natural Language Processing, pages 443-450, Hissar, Bulgaria, September. INCOMA Ltd. Shoumen, BULGARIA.", "links": null }, "BIBREF29": { "ref_id": "b29", "title": "Modeling trolling in social media conversations", "authors": [ { "first": "", "middle": [], "last": "Mojica De La", "suffix": "" }, { "first": "L", "middle": [ "G" ], "last": "Vega", "suffix": "" }, { "first": "V", "middle": [], "last": "Ng", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mojica de la Vega, L. G. and Ng, V. (2018). Modeling trolling in social media conversations. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan, May. European Language Resources Association (ELRA).", "links": null }, "BIBREF30": { "ref_id": "b30", "title": "Abusive language detection in online user content", "authors": [ { "first": "C", "middle": [], "last": "Nobata", "suffix": "" }, { "first": "J", "middle": [], "last": "Tetreault", "suffix": "" }, { "first": "A", "middle": [], "last": "Thomas", "suffix": "" }, { "first": "Y", "middle": [], "last": "Mehdad", "suffix": "" }, { "first": "Chang", "middle": [], "last": "", "suffix": "" }, { "first": "Y", "middle": [], "last": "", "suffix": "" } ], "year": 2016, "venue": "Proceedings of the 25th international conference on world wide web", "volume": "", "issue": "", "pages": "145--153", "other_ids": {}, "num": null, "urls": [], "raw_text": "Nobata, C., Tetreault, J., Thomas, A., Mehdad, Y., and Chang, Y. (2016). Abusive language detection in online user content. In Proceedings of the 25th international conference on world wide web, pages 145-153.", "links": null }, "BIBREF31": { "ref_id": "b31", "title": "Fighting offensive language on social media with unsupervised text style transfer", "authors": [ { "first": "C", "middle": [], "last": "Nogueira Dos Santos", "suffix": "" }, { "first": "I", "middle": [], "last": "Melnyk", "suffix": "" }, { "first": "I", "middle": [], "last": "Padhi", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics", "volume": "2", "issue": "", "pages": "189--194", "other_ids": {}, "num": null, "urls": [], "raw_text": "Nogueira dos Santos, C., Melnyk, I., and Padhi, I. (2018). Fighting offensive language on social media with unsu- pervised text style transfer. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 189-194, Melbourne, Australia, July. Association for Computa- tional Linguistics.", "links": null }, "BIBREF32": { "ref_id": "b32", "title": "Named entity recognition for code-mixed Indian corpus using meta embedding", "authors": [ { "first": "R", "middle": [], "last": "Priyadharshini", "suffix": "" }, { "first": "B", "middle": [ "R" ], "last": "Chakravarthi", "suffix": "" }, { "first": "M", "middle": [], "last": "Vegupatti", "suffix": "" }, { "first": "J", "middle": [ "P" ], "last": "Mccrae", "suffix": "" } ], "year": 2020, "venue": "2020 6th International Conference on Advanced Computing and Communication Systems (ICACCS)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Priyadharshini, R., Chakravarthi, B. R., Vegupatti, M., and McCrae, J. P. (2020). Named entity recognition for code-mixed Indian corpus using meta embedding. In 2020 6th International Conference on Advanced Com- puting and Communication Systems (ICACCS).", "links": null }, "BIBREF33": { "ref_id": "b33", "title": "A comparative study of different state-of-the-art hate speech detection methods for Hindi-English code-mixed data", "authors": [ { "first": "P", "middle": [], "last": "Rani", "suffix": "" }, { "first": "S", "middle": [], "last": "Suryawanshi", "suffix": "" }, { "first": "K", "middle": [], "last": "Goswami", "suffix": "" }, { "first": "B", "middle": [ "R" ], "last": "Chakravarthi", "suffix": "" }, { "first": "T", "middle": [], "last": "Fransen", "suffix": "" }, { "first": "J", "middle": [ "P" ], "last": "Mccrae", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the Second Workshop on Trolling, Aggression and Cyberbullying", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Rani, P., Suryawanshi, S., Goswami, K., Chakravarthi, B. R., Fransen, T., and McCrae, J. P. (2020). A compar- ative study of different state-of-the-art hate speech de- tection methods for Hindi-English code-mixed data. In Proceedings of the Second Workshop on Trolling, Ag- gression and Cyberbullying, Marseille, France, May. Eu- ropean Language Resources Association (ELRA).", "links": null }, "BIBREF34": { "ref_id": "b34", "title": "A comparative study on code-mixed data of Indian social media vs formal text", "authors": [ { "first": "P", "middle": [], "last": "Ranjan", "suffix": "" }, { "first": "B", "middle": [], "last": "Raja", "suffix": "" }, { "first": "R", "middle": [], "last": "Priyadharshini", "suffix": "" }, { "first": "R", "middle": [ "C" ], "last": "Balabantaray", "suffix": "" } ], "year": 2016, "venue": "2016 2nd International Conference on Contemporary Computing and Informatics (IC3I)", "volume": "", "issue": "", "pages": "608--611", "other_ids": {}, "num": null, "urls": [], "raw_text": "Ranjan, P., Raja, B., Priyadharshini, R., and Balabantaray, R. C. (2016). A comparative study on code-mixed data of Indian social media vs formal text. In 2016 2nd Inter- national Conference on Contemporary Computing and Informatics (IC3I), pages 608-611, Dec.", "links": null }, "BIBREF35": { "ref_id": "b35", "title": "Tamil English cross lingual information retrieval", "authors": [ { "first": "T", "middle": [ "P R K" ], "last": "Rao", "suffix": "" }, { "first": "S", "middle": [], "last": "Devi", "suffix": "" } ], "year": 2013, "venue": "Multilingual Information Access in South Asian Languages", "volume": "", "issue": "", "pages": "269--279", "other_ids": {}, "num": null, "urls": [], "raw_text": "Rao, T. P. R. K. and Lalitha Devi, S. (2013). Tamil En- glish cross lingual information retrieval. In Prasenjit Ma- jumder, et al., editors, Multilingual Information Access in South Asian Languages, pages 269-279, Berlin, Hei- delberg. Springer Berlin Heidelberg.", "links": null }, "BIBREF36": { "ref_id": "b36", "title": "ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision (IJCV)", "authors": [ { "first": "O", "middle": [], "last": "Russakovsky", "suffix": "" }, { "first": "J", "middle": [], "last": "Deng", "suffix": "" }, { "first": "H", "middle": [], "last": "Su", "suffix": "" }, { "first": "J", "middle": [], "last": "Krause", "suffix": "" }, { "first": "S", "middle": [], "last": "Satheesh", "suffix": "" }, { "first": "S", "middle": [], "last": "Ma", "suffix": "" }, { "first": "Z", "middle": [], "last": "Huang", "suffix": "" }, { "first": "A", "middle": [], "last": "Karpathy", "suffix": "" }, { "first": "A", "middle": [], "last": "Khosla", "suffix": "" }, { "first": "M", "middle": [], "last": "Bernstein", "suffix": "" }, { "first": "A", "middle": [ "C" ], "last": "Berg", "suffix": "" }, { "first": "L", "middle": [], "last": "Fei-Fei", "suffix": "" } ], "year": 2015, "venue": "", "volume": "115", "issue": "", "pages": "211--252", "other_ids": {}, "num": null, "urls": [], "raw_text": "Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., Berg, A. C., and Fei-Fei, L. (2015). ImageNet Large Scale Visual Recognition Challenge. International Jour- nal of Computer Vision (IJCV), 115(3):211-252.", "links": null }, "BIBREF37": { "ref_id": "b37", "title": "Multimodal meme dataset (Multi-OFF) for identifying offensive content in image and text", "authors": [ { "first": "S", "middle": [], "last": "Suryawanshi", "suffix": "" }, { "first": "B", "middle": [ "R" ], "last": "Chakravarthi", "suffix": "" }, { "first": "M", "middle": [], "last": "Arcan", "suffix": "" }, { "first": "P", "middle": [], "last": "Buitelaar", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the Second Workshop on Trolling, Aggression and Cyberbullying", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Suryawanshi, S., Chakravarthi, B. R., Arcan, M., and Buitelaar, P. (2020). Multimodal meme dataset (Multi- OFF) for identifying offensive content in image and text. In Proceedings of the Second Workshop on Trolling, Ag- gression and Cyberbullying, Marseille, France, May. Eu- ropean Language Resources Association (ELRA).", "links": null }, "BIBREF38": { "ref_id": "b38", "title": "I can has cheezburger? a nonparanormal approach to combining textual and visual information for predicting and generating popular meme descriptions", "authors": [ { "first": "W", "middle": [ "Y" ], "last": "Wang", "suffix": "" }, { "first": "M", "middle": [], "last": "Wen", "suffix": "" } ], "year": 2015, "venue": "Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", "volume": "", "issue": "", "pages": "355--365", "other_ids": {}, "num": null, "urls": [], "raw_text": "Wang, W. Y. and Wen, M. (2015). I can has cheezburger? a nonparanormal approach to combining textual and visual information for predicting and generating popular meme descriptions. In Proceedings of the 2015 Conference of the North American Chapter of the Association for Com- putational Linguistics: Human Language Technologies, pages 355-365, Denver, Colorado, May-June. Associa- tion for Computational Linguistics.", "links": null }, "BIBREF39": { "ref_id": "b39", "title": "Detecting offensive tweets via topical feature discovery over a large scale twitter corpus", "authors": [ { "first": "G", "middle": [], "last": "Xiang", "suffix": "" }, { "first": "B", "middle": [], "last": "Fan", "suffix": "" }, { "first": "L", "middle": [], "last": "Wang", "suffix": "" }, { "first": "J", "middle": [], "last": "Hong", "suffix": "" }, { "first": "Rose", "middle": [], "last": "", "suffix": "" }, { "first": "C", "middle": [], "last": "", "suffix": "" } ], "year": 2012, "venue": "Proceedings of the 21st ACM international conference on Information and knowledge management", "volume": "", "issue": "", "pages": "1980--1984", "other_ids": {}, "num": null, "urls": [], "raw_text": "Xiang, G., Fan, B., Wang, L., Hong, J., and Rose, C. (2012). Detecting offensive tweets via topical feature discovery over a large scale twitter corpus. In Proceed- ings of the 21st ACM international conference on Infor- mation and knowledge management, pages 1980-1984.", "links": null }, "BIBREF40": { "ref_id": "b40", "title": "From image descriptions to visual denotations: New similarity metrics for semantic inference over event descriptions", "authors": [ { "first": "P", "middle": [], "last": "Young", "suffix": "" }, { "first": "A", "middle": [], "last": "Lai", "suffix": "" }, { "first": "M", "middle": [], "last": "Hodosh", "suffix": "" }, { "first": "J", "middle": [], "last": "Hockenmaier", "suffix": "" } ], "year": 2014, "venue": "TACL", "volume": "2", "issue": "", "pages": "67--78", "other_ids": {}, "num": null, "urls": [], "raw_text": "Young, P., Lai, A., Hodosh, M., and Hockenmaier, J. (2014). From image descriptions to visual denotations: New similarity metrics for semantic inference over event descriptions. TACL, 2:67-78.", "links": null } }, "ref_entries": { "FIGREF0": { "uris": null, "num": null, "text": "Examples of Indian memes.", "type_str": "figure" }, "FIGREF1": { "uris": null, "num": null, "text": "Examples of troll and not-troll memes.", "type_str": "figure" }, "FIGREF2": { "uris": null, "num": null, "text": "Examples on same image with different text.", "type_str": "figure" }, "TABREF0": { "num": null, "html": null, "type_str": "table", "content": "
VariationsTamilMemesTamilMemes + ImageNet
Precision Recall f1-score count Precision Recall f1-scorecount
troll0.370.330.351000.360.350.35100
not-troll0.680.710.702000.680.690.68200
macro-avg0.520.520.523000.520.520.52300
weighted-avg0.580.590.583000.570.570.57300
VariationsTamilMemes + ImageNet + Flickr1k TamilMemes + ImageNet + Flickr30k
troll0.300.340.321000.360.350.35100
not-troll0.640.590.622000.680.690.68200
macro-avg0.470.470.473000.520.520.52300
weighted-avg0.530.510.523000.570.570.57300
MobileNet
VariationsTamilMemesTamilMemes + ImageNet
troll0.280.270.281000.340.430.38100
not-troll0.640.660.652000.670.580.62200
macro-avg0.460.460.463000.500.510.50300
weighted-avg0.520.530.523000.560.530.54300
VariationsTamilMemes + ImageNet + Flickr1k TamilMemes + ImageNet + Flickr30k
troll0.330.550.411000.310.340.33100
not-troll0.660.450.532000.650.620.64200
macro-avg0.500.500.473000.480.480.48300
weighted-avg0.550.480.493000.540.530.53300
", "text": "Precision, recall, F1-score and count for ResNet, MobileNet and their variations." } } } }