Risch, J., Ruff, R., Krestel, R.: Explaining Offensive Language Detection.Journal for Language Technology and Computational Linguistics (JLCL).34,1-19 (2020).
Weitere Informationen
HerausgeberRuppenhofer, Josef and Siegel, Melanie and Struß, Julia Maria
AbstractMachine learning approaches have proven to be on or even above human-level accuracy for the task of offensive language detection. In contrast to human experts, however, they often lack the capability of giving explanations for their decisions. This article compares four different approaches to make offensive language detection explainable: an interpretable machine learning model (naive Bayes), a model-agnostic explainability method (LIME), a model-based explainability method (LRP), and a self-explanatory model (LSTM with an attention mechanism). Three different classification methods: SVM, naive Bayes, and LSTM are paired with appropriate explanation methods. To this end, we investigate the trade-off between classification performance and explainability of the respective classifiers. We conclude that, with the appropriate explanation methods, the superior classification performance of more complex models is worth the initial lack of explainability.
Risch, J., Garda, S., Krestel, R.: Hierarchical Document Classification as a Sequence Generation Task.Proceedings of the Joint Conference on Digital Libraries (JCDL) (2020).
Weitere Informationen
AbstractHierarchical classification schemes are an effective and natural way to organize large document collections. However, complex schemes make the manual classification time-consuming and require domain experts. Current machine learning approaches for hierarchical classification do not exploit all the information contained in the hierarchical schemes. During training, they do not make full use of the inherent parent-child relation of classes. For example, they neglect to tailor document representations, such as embeddings, to each individual hierarchy level. Our model overcomes these problems by addressing hierarchical classification as a sequence generation task. To this end, our neural network transforms a sequence of input words into a sequence of labels, which represents a path through a tree-structured hierarchy scheme. The evaluation uses a patent corpus, which exhibits a complex class hierarchy scheme and high-quality annotations from domain experts and comprises millions of documents. We re-implemented five models from related work and show that our basic model achieves competitive results in comparison with the best approach. A variation of our model that uses the recent Transformer architecture outperforms the other approaches. The error analysis reveals that the encoder of our model has the strongest influence on its classification performance.
Risch, J., Krestel, R.: Bagging BERT Models for Robust Aggression Identification.Proceedings of the Workshop on Trolling, Aggression and Cyberbullying (TRAC@LREC). bll. 55-61. European Language Resources Association (ELRA) (2020).
Weitere Informationen
AbstractModern transformer-based models with hundreds of millions of parameters, such as BERT, achieve impressive results at text classification tasks. This also holds for aggression identification and offensive language detection, where deep learning approaches consistently outperform less complex models, such as decision trees. While the complex models fit training data well (low bias), they also come with an unwanted high variance. Especially when fine-tuning them on small datasets, the classification performance varies significantly for slightly different training data. To overcome the high variance and provide more robust predictions, we propose an ensemble of multiple fine-tuned BERT models based on bootstrap aggregating (bagging). In this paper, we describe such an ensemble system and present our submission to the shared tasks on aggression identification 2020 (team name: Julian). Our submission is the best-performing system for five out of six subtasks. For example, we achieve a weighted F1-score of 80.3% for task A on the test dataset of English social media posts. In our experiments, we compare different model configurations and vary the number of models used in the ensemble. We find that the F1-score drastically increases when ensembling up to 15 models, but the returns diminish for more models.
Risch, J., Krestel, R.: Top Comment or Flop Comment? Predicting and Explaining User Engagement in Online News Discussions.Proceedings of the International Conference on Web and Social Media (ICWSM). bll. 579-589. AAAI (2020).
Weitere Informationen
AbstractComment sections below online news articles enjoy growing popularity among readers. However, the overwhelming number of comments makes it infeasible for the average news consumer to read all of them and hinders engaging discussions. Most platforms display comments in chronological order, which neglects that some of them are more relevant to users and are better conversation starters. In this paper, we systematically analyze user engagement in the form of the upvotes and replies that a comment receives. Based on comment texts, we train a model to distinguish comments that have either a high or low chance of receiving many upvotes and replies. Our evaluation on user comments from TheGuardian.com compares recurrent and convolutional neural network models, and a traditional feature-based classifier. Further, we investigate what makes some comments more engaging than others. To this end, we identify engagement triggers and arrange them in a taxonomy. Explanation methods for neural networks reveal which input words have the strongest influence on our model's predictions. In addition, we evaluate on a dataset of product reviews, which exhibit similar properties as user comments, such as featuring upvotes for helpfulness.
Risch, J., Krestel, R.: Toxic Comment Detection in Online Discussions. In: Agarwal, B., Nayak, R., Mittal, N., en Patnaik, S. (reds.) Deep Learning-Based Approaches for Sentiment Analysis. bll. 85-109. Springer (2020).
Weitere Informationen
HerausgeberAgarwal, Basant and Nayak, Richi and Mittal, Namita and Patnaik, Srikanta
AbstractWith the exponential growth in the use of social media networks such as Twitter, Facebook, and many others, an astronomical amount of big data has been generated. A substantial amount of this user-generated data is in form of text such as reviews, tweets, and blogs that provide numerous challenges as well as opportunities to NLP (Natural Language Processing) researchers for discovering meaningful information used in various applications. Sentiment analysis is the study that analyses people’s opinion and sentiment towards entities such as products, services, person, organisations etc. present in the text. Sentiment analysis and opinion mining is the most popular and interesting research problem. In recent years, Deep Learning approaches have emerged as powerful computational models and have shown significant success to deal with a massive amount of data in unsupervised settings. Deep learning is revolutionizing because it offers an effective way of learning representation and allows the system to learn features automatically from data without the need of explicitly designing them. Deep learning algorithms such as deep autoencoders, convolutional and recurrent neural networks (CNN) (RNN), Long Short-Term Memory (LSTM) and Generative Adversarial Networks (GAN) have reported providing significantly improved results in various natural language processing tasks including sentiment analysis.
Risch, J., Ruff, R., Krestel, R.: Offensive Language Detection Explained.Proceedings of the Workshop on Trolling, Aggression and Cyberbullying (TRAC@LREC). bll. 137-143. European Language Resources Association (ELRA) (2020).
Weitere Informationen
AbstractMany online discussion platforms use a content moderation process, where human moderators check user comments for offensive language and other rule violations. It is the moderator's decision which comments to remove from the platform because of violations and which ones to keep. Research so far focused on automating this decision process in the form of supervised machine learning for a classification task. However, even with machine-learned models achieving better classification accuracy than human experts in some scenarios, there is still a reason why human moderators are preferred. In contrast to black-box models, such as neural networks, humans can give explanations for their decision to remove a comment. For example, they can point out which phrase in the comment is offensive or what subtype of offensiveness applies. In this paper, we analyze and compare four attribution-based explanation methods for different offensive language classifiers: an interpretable machine learning model (naive Bayes), a model-agnostic explanation method (LIME), a model-based explanation method (LRP), and a self-explanatory model (LSTM with an attention mechanism). We evaluate these approaches with regard to their explanatory power and their ability to point out which words are most relevant for a classifier's decision. We find that the more complex models achieve better classification accuracy while also providing better explanations than the simpler models.
Risch, J., Stoll, A., Ziegele, M., Krestel, R.: hpiDEDIS at GermEval 2019: Offensive Language Identification using a German BERT model.Proceedings of the 15th Conference on Natural Language Processing (KONVENS). bl. 403--408. German Society for Computational Linguistics & Language Technology, Erlangen, Germany (2019).
Weitere Informationen
AbstractPre-training language representations on large text corpora, for example, with BERT, has recently shown to achieve impressive performance at a variety of downstream NLP tasks. So far, applying BERT to offensive language identification for German- language texts failed due to the lack of pre-trained, German-language models. In this paper, we fine-tune a BERT model that was pre-trained on 12 GB of German texts to the task of offensive language identification. This model significantly outperforms our baselines and achieves a macro F1 score of 76% on coarse-grained, 51% on fine-grained, and 73% on implicit/explicit classification. We analyze the strengths and weaknesses of the model and derive promising directions for future work.
Risch, J., Krestel, R.: Measuring and Facilitating Data Repeatability in Web Science.Datenbank-Spektrum.19,117-126 (2019).
Weitere Informationen
AbstractAccessible and reusable datasets are a necessity to accomplish repeatable research. This requirement poses a problem particularly for web science, since scraped data comes in various formats and can change due to the dynamic character of the web. Further, usage of web data is typically restricted by copyright-protection or privacy regulations, which hinder publication of datasets. To alleviate these problems and reach what we define as “partial data repeatability”, we present a process that consists of multiple components. Researchers need to distribute only a scraper and not the data itself to comply with legal limitations. If a dataset is re-scraped for repeatability after some time, the integrity of different versions can be checked based on fingerprints. Moreover, fingerprints are sufficient to identify what parts of the data have changed and how much. We evaluate an implementation of this process with a dataset of 250 million online comments collected from five different news discussion platforms. We re-scraped the dataset after pausing for one year and show that less than ten percent of the data has actually changed. These experiments demonstrate that providing a scraper and fingerprints enables recreating a dataset and supports the repeatability of web science experiments.
Risch, J., Krestel, R.: Domain-specific word embeddings for patent classification.Data Technologies and Applications.53,108-122 (2019).
Weitere Informationen
AbstractPatent offices and other stakeholders in the patent domain need to classify patent applications according to a standardized classification scheme. To examine the novelty of an application it can then be compared to previously granted patents in the same class. Automatic classification would be highly beneficial, because of the large volume of patents and the domain-specific knowledge needed to accomplish this costly manual task. However, a challenge for the automation is patent-specific language use, such as special vocabulary and phrases. To account for this language use, we present domain-specific pre-trained word embeddings for the patent domain. We train our model on a very large dataset of more than 5 million patents and evaluate it at the task of patent classification. To this end, we propose a deep learning approach based on gated recurrent units for automatic patent classification built on the trained word embeddings. Experiments on a standardized evaluation dataset show that our approach increases average precision for patent classification by 17 percent compared to state-of-the-art approaches. In this paper, we further investigate the model’s strengths and weaknesses. An extensive error analysis reveals that the learned embeddings indeed mirror patent-specific language use. The imbalanced training data and underrepresented classes are the most difficult remaining challenge.
Risch, J., Krestel, R.: Delete or not Delete? Semi-Automatic Comment Moderation for the Newsroom.Proceedings of the First Workshop on Trolling, Aggression and Cyberbullying (co-located with COLING). bll. 166-176 (2018).
Weitere Informationen
AbstractComment sections of online news providers have enabled millions to share and discuss their opinions on news topics. Today, moderators ensure respectful and informative discussions by deleting not only insults, defamation, and hate speech, but also unverifiable facts. This process has to be transparent and comprehensive in order to keep the community engaged. Further, news providers have to make sure to not give the impression of censorship or dissemination of fake news. Yet manual moderation is very expensive and becomes more and more unfeasible with the increasing amount of comments. Hence, we propose a semi-automatic, holistic approach, which includes comment features but also their context, such as information about users and articles. For evaluation, we present experiments on a novel corpus of 3 million news comments annotated by a team of professional moderators.
Risch, J., Krestel, R.: My Approach = Your Apparatus? Entropy-Based Topic Modeling on Multiple Domain-Specific Text Collections.Proceedings of the 18th ACM/IEEE Joint Conference on Digital Libraries (JCDL). bll. 283-292 (2018).
Weitere Informationen
AbstractComparative text mining extends from genre analysis and political bias detection to the revelation of cultural and geographic differences, through to the search for prior art across patents and scientific papers. These applications use cross-collection topic modeling for the exploration, clustering, and comparison of large sets of documents, such as digital libraries. However, topic modeling on documents from different collections is challenging because of domain-specific vocabulary. We present a cross-collection topic model combined with automatic domain term extraction and phrase segmentation. This model distinguishes collection-specific and collection-independent words based on information entropy and reveals commonalities and differences of multiple text collections. We evaluate our model on patents, scientific papers, newspaper articles, forum posts, and Wikipedia articles. In comparison to state-of-the-art cross-collection topic modeling, our model achieves up to 13% higher topic coherence, up to 4% lower perplexity, and up to 31% higher document classification accuracy. More importantly, our approach is the first topic model that ensures disjunct general and specific word distributions, resulting in clear-cut topic representations.
Ambroselli, C., Risch, J., Krestel, R., Loos, A.: Prediction for the Newsroom: Which Articles Will Get the Most Comments?Proceedings of the 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics (NAACL). bll. 193-199. ACL, New Orleans, Louisiana, USA (2018).
Weitere Informationen
AbstractThe overwhelming success of the Web and mobile technologies has enabled millions to share their opinions publicly at any time. But the same success also endangers this freedom of speech due to closing down of participatory sites misused by individuals or interest groups. We propose to support manual moderation by proactively drawing the attention of our moderators to article discussions that most likely need their intervention. To this end, we predict which articles will receive a high number of comments. In contrast to existing work, we enrich the article with metadata, extract semantic and linguistic features, and exploit annotated data from a foreign language corpus. Our logistic regression model improves F1-scores by over 80% in comparison to state-of-the-art approaches.
Risch, J., Krestel, R.: Learning Patent Speak: Investigating Domain-Specific Word Embeddings.Proceedings of the Thirteenth International Conference on Digital Information Management (ICDIM). bll. 63-68 (2018).
Weitere Informationen
AbstractA patent examiner needs domain-specific knowledge to classify a patent application according to its field of invention. Standardized classification schemes help to compare a patent application to previously granted patents and thereby check its novelty. Due to the large volume of patents, automatic patent classification would be highly beneficial to patent offices and other stakeholders in the patent domain. However, a challenge for the automation of this costly manual task is the patent-specific language use. To facilitate this task, we present domain-specific pre-trained word embeddings for the patent domain. We trained our model on a very large dataset of more than 5 million patents to learn the language use in this domain. We evaluated the quality of the resulting embeddings in the context of patent classification. To this end, we propose a deep learning approach based on gated recurrent units for automatic patent classification built on the trained word embeddings. Experiments on a standardized evaluation dataset show that our approach increases average precision for patent classification by 17 percent compared to state-of-the-art approaches.
Risch, J., Krestel, R.: Aggression Identification Using Deep Learning and Data Augmentation.Proceedings of the First Workshop on Trolling, Aggression and Cyberbullying (co-located with COLING). bll. 150-158 (2018).
Weitere Informationen
AbstractSocial media platforms allow users to share and discuss their opinions online. However, a minority of user posts is aggressive, thereby hinders respectful discussion, and — at an extreme level — is liable to prosecution. The automatic identification of such harmful posts is important, because it can support the costly manual moderation of online discussions. Further, the automation allows unprecedented analyses of discussion datasets that contain millions of posts. This system description paper presents our submission to the First Shared Task on Aggression Identification. We propose to augment the provided dataset to increase the number of labeled comments from 15,000 to 60,000. Thereby, we introduce linguistic variety into the dataset. As a consequence of the larger amount of training data, we are able to train a special deep neural net, which generalizes especially well to unseen data. To further boost the performance, we combine this neural net with three logistic regression classifiers trained on character and word n-grams, and hand-picked syntactic features. This ensemble is more robust than the individual single models. Our team named “Julian” achieves an F1-score of 60% on both English datasets, 63% on the Hindi Facebook dataset, and 38% on the Hindi Twitter dataset.
van Aken, B., Risch, J., Krestel, R., Löser, A.: Challenges for Toxic Comment Classification: An In-Depth Error Analysis.Proceedings of the 2nd Workshop on Abusive Language Online (co-located with EMNLP). bll. 33-42 (2018).
Weitere Informationen
AbstractToxic comment classification has become an active research field with many recently proposed approaches. However, while these approaches address some of the task’s challenges others still remain unsolved and directions for further research are needed. To this end, we compare different approaches on a new, large comment dataset and propose an ensemble that outperforms all individual models. Further, we validate our findings on a second dataset. The results of the ensemble enable us to perform an extensive error analysis, which reveals open challenges for state-of- the-art methods and directions towards pending future research. These challenges include missing paradigmatic context and inconsistent dataset labels.
Risch, J., Krebs, E., Löser, A., Riese, A., Krestel, R.: Fine-Grained Classification of Offensive Language.Proceedings of GermEval (co-located with KONVENS). bll. 38-44 (2018).
Weitere Informationen
AbstractSocial media platforms receive massive amounts of user-generated content that may include offensive text messages. In the context of the GermEval task 2018, we propose an approach for fine-grained classification of offensive language. Our approach comprises a Naive Bayes classifier, a neural network, and a rule-based approach that categorize tweets. In addition, we combine the approaches in an ensemble to overcome weaknesses of the single models. We cross-validate our approaches with regard to macro-average F1-score on the provided training dataset.
Risch, J., Garda, S., Krestel, R.: Book Recommendation Beyond the Usual Suspects: Embedding Book Plots Together with Place and Time Information.Proceedings of the 20th International Conference On Asia-Pacific Digital Libraries (ICADL). bll. 227-239 (2018).
Weitere Informationen
AbstractContent-based recommendation of books and other media is usually based on semantic similarity measures. While metadata can be compared easily, measuring the semantic similarity of narrative literature is challenging. Keyword-based approaches are biased to retrieve books of the same series or do not retrieve any results at all in sparser libraries. We propose to represent plots with dense vectors to foster semantic search for similar plots even if they do not have any words in common. Further, we propose to embed plots, places, and times in the same embedding space. Thereby, we allow arithmetics on these aspects. For example, a book with a similar plot but set in a different, user-specified place can be retrieved. We evaluate our findings on a set of 16,000 book synopses that spans literature from 500 years and 200 genres and compare our approach to a keyword-based baseline.
Maschler, F., Niephaus, F., Risch, J.: Real or Fake? Large-Scale Validation of Identity Leaks.47. Jahrestagung der Gesellschaft für Informatik (INFORMATIK). bll. 2437-2448 (2017).
Weitere Informationen
AbstractOn the Internet, criminal hackers frequently leak identity data on a massive scale. Subsequent criminal activities, such as identity theft and misuse, put Internet users at risk. Leak checker services enable users to check whether their personal data has been made public. However, automatic crawling and identification of leak data is error-prone for different reasons. Based on a dataset of more than 180 million leaked identity records, we propose a software system that identifies and validates identity leaks to improve leak checker services. Furthermore, we present a proficient assessment of leak data quality and typical characteristics that distinguish valid and invalid leaks.
Krestel, R., Risch, J.: How Do Search Engines Work? A Massive Open Online Course with 4000 Participants.Proceedings of the Conference Lernen, Wissen, Daten, Analysen. bll. 259-271 (2017).
Weitere Informationen
AbstractMassive Open Online Courses (MOOCs) have introduced a new form of education. With thousands of participants per course, lectur- ers are confronted with new challenges in the teaching process. In this pa- per, we describe how we conducted an introductory information retrieval course for participants from all ages and educational backgrounds. We analyze different course phases and compare our experiences with regular on-site information retrieval courses at university.
Risch, J., Krestel, R.: What Should I Cite? Cross-Collection Reference Recommendation of Patents and Papers.Proceedings of the International Conference on Theory and Practice of Digital Libraries (TPDL). bll. 40-46 (2017).
Weitere Informationen
AbstractResearch results manifest in large corpora of patents and scientific papers. However, both corpora lack a consistent taxonomy and references across different document types are sparse. Therefore, and because of contrastive, domain-specific language, recommending similar papers for a given patent (or vice versa) is challenging. We propose a hybrid recommender system that leverages topic distributions and key terms to recommend related work despite these challenges. As a case study, we evaluate our approach on patents and papers of two fields: medical and computer science. We find that topic-based recommenders complement term-based recommenders for documents with collection-specific language and increase mean average precision by up to 23%. As a result of our work, publications from both corpora form a joint digital library, which connects academia and industry.
Bleifuß, T., Bülow, S., Frohnhofen, J., Risch, J., Wiese, G., Kruse, S., Papenbrock, T., Naumann, F.: Approximate Discovery of Functional Dependencies for Large Datasets.Proceedings of the International Conference on Information and Knowledge Management (CIKM). bll. 1803-1812. ACM, New York, NY, USA (2016).
Weitere Informationen
AbstractFunctional dependencies (FDs) are an important prerequisite for various data management tasks, such as schema normalization, query optimization, and data cleansing. However, automatic FD discovery entails an exponentially growing search and solution space, so that even today’s fastest FD discovery algorithms are limited to small datasets only, due to long runtimes and high memory consumptions. To overcome this situation, we propose an approximate discovery strategy that sacrifices possibly little result correctness in return for large performance improvements. In particular, we introduce AID-FD, an algorithm that approximately discovers FDs within runtimes up to orders of magnitude faster than state-of-the-art FD discovery algorithms. We evaluate and compare our performance results with a focus on scalability in runtime and memory, and with measures for completeness, correctness, and minimality.
Hennig, P., Berger, P., Dullweber, C., Finke, M., Maschler, F., Risch, J., Meinel, C.: Social Media Story Telling.Proceedings of the 8th IEEE International Conference on Social Computing and Networking (SocialCom2015). bll. 279-284. , Chengdu, China (2015).
Weitere Informationen
AbstractThe number of documents on the web increases rapidly and often there is an enormous information overlap between different sources covering the same topic. Since it is impractical to read through all posts regarding a subject, there is a need for summaries combining the most relevant facts. In this context combining information from different sources in form of stories is an important method to provide perspective, while presenting and enriching the existing content in an interesting, natural and narrative way. Today, stories are often not available or they have been elaborately written and selected by journalists. Thus, we present an automated approach to create stories from multiple input documents. Furthermore the developed framework implements strategies to visualize stories and link content to related sources of information, such as images, tweets and encyclopedia records ready to be explored by the reader. Our approach combines deriving a story line from a graph of interlinked sources with a story-centric multi-document summarization.
Schmidt, D., Frohnhofen, J., Knebel, S., Meinel, F., Perchyk, M., Risch, J., Striebel, J., Wachtel, J., Baudisch, P.: Ergonomic Interaction for Touch Floors.Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems. bll. 3879-3888. ACM, Seoul, Republic of Korea (2015).
Weitere Informationen
AbstractThe main appeal of touch floors is that they are the only direct touch form factor that scales to arbitrary size, therefore allowing direct touch to scale to very large numbers of display objects. In this paper, however, we argue that the price for this benefit is bad physical ergonomics: prolonged standing, especially in combination with looking down, quickly causes fatigue and repetitive strain. We propose addressing this issue by allowing users to operate touch floors in any pose they like, including sitting and lying. To allow users to transition between poses seamlessly, we present a simple pose-aware view manager that supports users by adjusting the entire view to the new pose. We support the main assumption behind the work with a simple study that shows that several poses are indeed more ergonomic for touch floor interaction than standing. We ground the design of our view manager by analyzing, which screen regions users can see and touch in each of the respective poses.
