Last Year

%0 Conference Proceedings %A Arnaout, Hiba %A Tran, Trung-Kien %A Stepanova, Daria %A Gad-Elrab, Mohamed Hassan %A Razniewski, Simon %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations External Organizations External Organizations External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Utilizing Language Model Probes for Knowledge Graph Repair : %G eng %U http://hdl.handle.net/21.11116/0000-000A-63F4-3 %U https://wikiworkshop.org/2022/ %D 2022 %B Wiki Workshop 2022 %Z date of event: 2022-04-25 - 2022-04-25 %C Virtual Event %B Wiki Workshop 2022

%0 Conference Proceedings %A Arnaout, Hiba %A Razniewski, Simon %A Weikum, Gerhard %A Pan, Jeff Z. %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations %T UnCommonSense: Informative Negative Knowledge about Everyday Concepts : %G eng %U http://hdl.handle.net/21.11116/0000-000A-F224-C %R 10.1145/3511808.3557484 %D 2022 %B 31st ACM International Conference on Information and Knowledge Management %Z date of event: 2022-10-17 - 2022-10-21 %C Atlanta GA USA %X Commonsense knowledge about everyday concepts is an important asset for AI<br>applications, such as question answering and chatbots. Recently, we have seen<br>an increasing interest in the construction of structured commonsense knowledge<br>bases (CSKBs). An important part of human commonsense is about properties that<br>do not apply to concepts, yet existing CSKBs only store positive statements.<br>Moreover, since CSKBs operate under the open-world assumption, absent<br>statements are considered to have unknown truth rather than being invalid. This<br>paper presents the UNCOMMONSENSE framework for materializing informative<br>negative commonsense statements. Given a target concept, comparable concepts<br>are identified in the CSKB, for which a local closed-world assumption is<br>postulated. This way, positive statements about comparable concepts that are<br>absent for the target concept become seeds for negative statement candidates.<br>The large set of candidates is then scrutinized, pruned and ranked by<br>informativeness. Intrinsic and extrinsic evaluations show that our method<br>significantly outperforms the state-of-the-art. A large dataset of informative<br>negations is released as a resource for future research.<br> %K Computer Science, Artificial Intelligence, cs.AI,Computer Science, Databases, cs.DB,Computer Science, Information Retrieval, cs.IR %B CIKM '22 %E Al Hasan, Mohammad; Xiong, Li %P 37 - 46 %I ACM %@ 978-1-4503-9236-5

%0 Report %A Arnaout, Hiba %A Razniewski, Simon %A Weikum, Gerhard %A Pan, Jeff Z. %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations %T UnCommonSense: Informative Negative Knowledge about Everyday Concepts : %G eng %U http://hdl.handle.net/21.11116/0000-000C-1651-0 %U https://arxiv.org/abs/2208.09292 %D 2022 %X Commonsense knowledge about everyday concepts is an important asset for AI<br>applications, such as question answering and chatbots. Recently, we have seen<br>an increasing interest in the construction of structured commonsense knowledge<br>bases (CSKBs). An important part of human commonsense is about properties that<br>do not apply to concepts, yet existing CSKBs only store positive statements.<br>Moreover, since CSKBs operate under the open-world assumption, absent<br>statements are considered to have unknown truth rather than being invalid. This<br>paper presents the UNCOMMONSENSE framework for materializing informative<br>negative commonsense statements. Given a target concept, comparable concepts<br>are identified in the CSKB, for which a local closed-world assumption is<br>postulated. This way, positive statements about comparable concepts that are<br>absent for the target concept become seeds for negative statement candidates.<br>The large set of candidates is then scrutinized, pruned and ranked by<br>informativeness. Intrinsic and extrinsic evaluations show that our method<br>significantly outperforms the state-of-the-art. A large dataset of informative<br>negations is released as a resource for future research.<br> %K Computer Science, Artificial Intelligence, cs.AI,Computer Science, Information Retrieval, cs.IR

%0 Conference Proceedings %A Chernyavsky, Irina %A Varde, Aparna S. %A Razniewski, Simon %+ External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T CSK-Detector: Commonsense in Object Detection : %G eng %U http://hdl.handle.net/21.11116/0000-000C-B77F-8 %R 10.1109/BigData55660.2022.10020915 %D 2022 %B IEEE International Conference on Big Data %Z date of event: 2022-12-17 - 2022-12-20 %C Osaka, Japan %B IEEE International Conference on Big Data %P 6609 - 6612 %I IEEE %@ 978-1-6654-8045-1

%0 Conference Proceedings %A Christmann, Philipp %A Saha Roy, Rishiraj %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Conversational Question Answering on Heterogeneous Sources : %G eng %U http://hdl.handle.net/21.11116/0000-000A-6148-8 %R 10.1145/3477495.3531815 %D 2022 %B 45th International ACM SIGIR Conference on Research and Development in Information Retrieval %Z date of event: 2022-07-11 - 2022-07-15 %C Madrid, Spain %B SIGIR '22 %E Amigo, Enrique; Castells, Pablo; Gonzalo, Julio; Carterett, Ben; Culpepper, J. Shane; Kazai, Gabriella %P 144 - 154 %I ACM %@ 978-1-4503-8732-3

%0 Conference Proceedings %A Christmann, Philipp %A Saha Roy, Rishiraj %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Beyond NED: Fast and Effective Search Space Reduction for Complex Question Answering over Knowledge Bases : %G eng %U http://hdl.handle.net/21.11116/0000-000A-27C6-B %R 10.1145/3488560.3498488 %D 2022 %B Fifteenth ACM International Conference on Web Search and Data Mining %Z date of event: 2022-02-21 - 2022-02-25 %C Tempe, AZ, USA (Virutal Event) %B WSDM '22 %P 172 - 180 %I ACM %@ 978-1-4503-9132-0

%0 Conference Proceedings %A Christmann, Philipp %A Saha Roy, Rishiraj %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Question Entity and Relation Linking to Knowledge Bases via CLOCQ : Question Entity and Relation Linking to Knowledge Bases via {CLOCQ} %G eng %U http://hdl.handle.net/21.11116/0000-000C-9595-3 %U https://ceur-ws.org/Vol-3337/smart-paper1.pdf %D 2022 %B 2nd Semantic Reasoning Evaluation Challenge 3rd SeMantic Answer Type, Relation and Entity Prediction Tasks Challengeand %Z date of event: 2022-10-24 - 2022-10-27 %C Hybrid Event, Hanghzou, China %B Joint Proceedings of SemREC 2022 and SMART 2022 co-located with 21st International Semantic Web Conference (ISWC 2022) %E Singh, Gunjan; Mutharaju, Raghava; Kapanipathi, Pavan; Mihindukulasooriya, Nandana; Dubey, Mohnish; Usbeck, Ricardo; Banerjee, Debayan %P 33 - 47 %I CEUR-WS.org %B CEUR Workshop Proceedings %N 3337

%0 Report %A Christmann, Philipp %A Saha Roy, Rishiraj %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Conversational Question Answering on Heterogeneous Sources : %G eng %U http://hdl.handle.net/21.11116/0000-000C-164E-5 %U https://arxiv.org/abs/2204.11677 %D 2022 %X Conversational question answering (ConvQA) tackles sequential information<br>needs where contexts in follow-up questions are left implicit. Current ConvQA<br>systems operate over homogeneous sources of information: either a knowledge<br>base (KB), or a text corpus, or a collection of tables. This paper addresses<br>the novel issue of jointly tapping into all of these together, this way<br>boosting answer coverage and confidence. We present CONVINSE, an end-to-end<br>pipeline for ConvQA over heterogeneous sources, operating in three stages: i)<br>learning an explicit structured representation of an incoming question and its<br>conversational context, ii) harnessing this frame-like representation to<br>uniformly capture relevant evidences from KB, text, and tables, and iii)<br>running a fusion-in-decoder model to generate the answer. We construct and<br>release the first benchmark, ConvMix, for ConvQA over heterogeneous sources,<br>comprising 3000 real-user conversations with 16000 questions, along with entity<br>annotations, completed question utterances, and question paraphrases.<br>Experiments demonstrate the viability and advantages of our method, compared to<br>state-of-the-art baselines.<br> %K Computer Science, Information Retrieval, cs.IR,Computer Science, Computation and Language, cs.CL

%0 Thesis %A Chu, Cuong Xuan %Y Weikum, Gerhard %A referee: Theobald, Martin %+ Databases and Information Systems, MPI for Informatics, Max Planck Society International Max Planck Research School, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Knowledge Extraction from Fictional Texts : %G eng %U http://hdl.handle.net/21.11116/0000-000A-9598-2 %R 10.22028/D291-36107 %U nbn:de:bsz:291--ds-361070 %F OTHER: hdl:20.500.11880/32914 %I Universit&#228;t des Saarlandes %C Saarbr&#252;cken %D 2022 %P 129 p. %V phd %9 phd %X Knowledge extraction from text is a key task in natural language processing, which involves many sub-tasks, such as taxonomy induction, named entity recognition and typing, relation extraction, knowledge canonicalization and so on. By constructing structured knowledge from natural language text, knowledge extraction becomes a key asset for search engines, question answering and other downstream applications. However, current knowledge extraction methods mostly focus on prominent real-world entities with Wikipedia and mainstream news articles as sources. The constructed knowledge bases, therefore, lack information about long-tail domains, with fiction and fantasy as archetypes. Fiction and fantasy are core parts of our human culture, spanning from literature to movies, TV series, comics and video games. With thousands of fictional universes which have been created, knowledge from fictional domains are subject of search-engine queries - by fans as well as cultural analysts. Unlike the real-world domain, knowledge extraction on such specific domains like fiction and fantasy has to tackle several key challenges: - Training data: Sources for fictional domains mostly come from books and fan-built content, which is sparse and noisy, and contains difficult structures of texts, such as dialogues and quotes. Training data for key tasks such as taxonomy induction, named entity typing or relation extraction are also not available. - Domain characteristics and diversity: Fictional universes can be highly sophisticated, containing entities, social structures and sometimes languages that are completely different from the real world. State-of-the-art methods for knowledge extraction make assumptions on entity-class, subclass and entity-entity relations that are often invalid for fictional domains. With different genres of fictional domains, another requirement is to transfer models across domains. - Long fictional texts: While state-of-the-art models have limitations on the input sequence length, it is essential to develop methods that are able to deal with very long texts (e.g. entire books), to capture multiple contexts and leverage widely spread cues. This dissertation addresses the above challenges, by developing new methodologies that advance the state of the art on knowledge extraction in fictional domains. - The first contribution is a method, called TiFi, for constructing type systems (taxonomy induction) for fictional domains. By tapping noisy fan-built content from online communities such as Wikia, TiFi induces taxonomies through three main steps: category cleaning, edge cleaning and top-level construction. Exploiting a variety of features from the original input, TiFi is able to construct taxonomies for a diverse range of fictional domains with high precision. - The second contribution is a comprehensive approach, called ENTYFI, for named entity recognition and typing in long fictional texts. Built on 205 automatically induced high-quality type systems for popular fictional domains, ENTYFI exploits the overlap and reuse of these fictional domains on unseen texts. By combining different typing modules with a consolidation stage, ENTYFI is able to do fine-grained entity typing in long fictional texts with high precision and recall. - The third contribution is an end-to-end system, called KnowFi, for extracting relations between entities in very long texts such as entire books. KnowFi leverages background knowledge from 142 popular fictional domains to identify interesting relations and to collect distant training samples. KnowFi devises a similarity-based ranking technique to reduce false positives in training samples and to select potential text passages that contain seed pairs of entities. By training a hierarchical neural network for all relations, KnowFi is able to infer relations between entity pairs across long fictional texts, and achieves gains over the best prior methods for relation extraction. %U https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/32914

%0 Journal Article %A Dave, Dev %A Celestino, Angelica %A Varde, Aparna S. %A Anu, Vaibhav %+ External Organizations External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations %T Management of Implicit Requirements Data in Large SRS Documents: Taxonomy and Techniques : %G eng %U http://hdl.handle.net/21.11116/0000-000A-F1AD-3 %7 2022 %D 2022 %J Sigmod Record %V 51 %N 2 %& 18 %P 18 - 29 %I Special Interest Group on the Management of Data %C New York, NY %@ false

%0 Thesis %A Fischer, Jonas %Y Vreeken, Jilles %A referee: Weikum, Gerhard %A referee: Parthasarathy, Srinivasan %+ Databases and Information Systems, MPI for Informatics, Max Planck Society International Max Planck Research School, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations %T More than the sum of its parts : pattern mining neural networks, and how they complement each other %G eng %U http://hdl.handle.net/21.11116/0000-000B-38BF-0 %R 10.22028/D291-37024 %U nbn:de:bsz:291--ds-370240 %F OTHER: hdl:20.500.11880/33893 %I Universit&#228;t des Saarlandes %C Saarbr&#252;cken %D 2022 %P 250 p. %V phd %9 phd %X In this thesis we explore pattern mining and deep learning. Often seen as orthogonal, we show that these fields complement each other and propose to combine them to gain from each other&#8217;s strengths. We, first, show how to efficiently discover succinct and non-redundant sets of patterns that provide insight into data beyond conjunctive statements. We leverage the interpretability of such patterns to unveil how and which information flows through neural networks, as well as what characterizes their decisions. Conversely, we show how to combine continuous optimization with pattern discovery, proposing a neural network that directly encodes discrete patterns, which allows us to apply pattern mining at a scale orders of magnitude larger than previously possible. Large neural networks are, however, exceedingly expensive to train for which &#8216;lottery tickets&#8217; &#8211; small, well-trainable sub-networks in randomly initialized neural networks &#8211; offer a remedy. We identify theoretical limitations of strong tickets and overcome them by equipping these tickets with the property of universal approximation. To analyze whether limitations in ticket sparsity are algorithmic or fundamental, we propose a framework to plant and hide lottery tickets. With novel ticket benchmarks we then conclude that the limitation is likely algorithmic, encouraging further developments for which our framework offers means to measure progress. %U https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/33893

%0 Conference Proceedings %A Ghosh, Shrestha %A Razniewski, Simon %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Answering Count Queries with Explanatory Evidence : %G eng %U http://hdl.handle.net/21.11116/0000-000A-9E36-8 %R 10.1145/3477495.3531870 %D 2022 %B 45th International ACM SIGIR Conference on Research and Development in Information Retrieval %Z date of event: 2022-07-11 - 2022-07-15 %C Madrid, Spain %B SIGIR '22 %E Amigo, Enrique; Castells, Pablo; Gonzalo, Julio; Carterett, Ben; Culpepper, J. Shane; Kazai, Gabriella %P 2415 - 2419 %I ACM %@ 978-1-4503-8732-3

%0 Report %A Ghosh, Shrestha %A Razniewski, Simon %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Answering Count Questions with Structured Answers from Text : %G eng %U http://hdl.handle.net/21.11116/0000-000B-1D84-0 %R 10.48550/arXiv.2209.07250 %D 2022 %X In this work we address the challenging case of answering count queries in<br>web search, such as ``number of songs by John Lennon''. Prior methods merely<br>answer these with a single, and sometimes puzzling number or return a ranked<br>list of text snippets with different numbers. This paper proposes a methodology<br>for answering count queries with inference, contextualization and explanatory<br>evidence. Unlike previous systems, our method infers final answers from<br>multiple observations, supports semantic qualifiers for the counts, and<br>provides evidence by enumerating representative instances. Experiments with a<br>wide variety of queries, including existing benchmark show the benefits of our<br>method, and the influence of specific parameter settings. Our code, data and an<br>interactive system demonstration are publicly available at<br>https://github.com/ghoshs/CoQEx and https://nlcounqer.mpi-inf.mpg.de/.<br> %K Computer Science, Information Retrieval, cs.IR

%0 Thesis %A Guimar&#227;es, Anna %Y Weikum, Gerhard %A referee: de Melo, Gerard %A referee: Yates, Andrew %+ Databases and Information Systems, MPI for Informatics, Max Planck Society International Max Planck Research School, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Data Science Methods for the Analysis of Controversial Social Media Discussions : %G eng %U http://hdl.handle.net/21.11116/0000-000A-CDF7-9 %R 10.22028/D291-36502 %U nbn:de:bsz:291--ds-365021 %F OTHER: hdl:20.500.11880/33161 %I Universit&#228;t des Saarlandes %C Saarbr&#252;cken %D 2022 %P 94 p. %V phd %9 phd %X Social media communities like Reddit and Twitter allow users to express their views on<br>topics of their interest, and to engage with other users who may share or oppose these views.<br>This can lead to productive discussions towards a consensus, or to contended debates, where<br>disagreements frequently arise.<br>Prior work on such settings has primarily focused on identifying notable instances of antisocial<br>behavior such as hate-speech and &#8220;trolling&#8221;, which represent possible threats to the health of<br>a community. These, however, are exceptionally severe phenomena, and do not encompass<br>controversies stemming from user debates, differences of opinions, and off-topic content, all<br>of which can naturally come up in a discussion without going so far as to compromise its<br>development.<br>This dissertation proposes a framework for the systematic analysis of social media discussions<br>that take place in the presence of controversial themes, disagreements, and mixed opinions from<br>participating users. For this, we develop a feature-based model to describe key elements of a<br>discussion, such as its salient topics, the level of activity from users, the sentiments it expresses,<br>and the user feedback it receives.<br>Initially, we build our feature model to characterize adversarial discussions surrounding<br>political campaigns on Twitter, with a focus on the factual and sentimental nature of their<br>topics and the role played by different users involved. We then extend our approach to Reddit<br>discussions, leveraging community feedback signals to define a new notion of controversy<br>and to highlight conversational archetypes that arise from frequent and interesting interaction<br>patterns. We use our feature model to build logistic regression classifiers that can predict future<br>instances of controversy in Reddit communities centered on politics, world news, sports, and<br>personal relationships. Finally, our model also provides the basis for a comparison of different<br>communities in the health domain, where topics and activity vary considerably despite their<br>shared overall focus. In each of these cases, our framework provides insight into how user<br>behavior can shape a community&#8217;s individual definition of controversy and its overall identity. %U https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/33161

%0 Conference Proceedings %A Hedderich, Michael A. %A Fischer, Jonas %A Klakow, Dietrich %A Vreeken, Jilles %+ External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations External Organizations %T Label-Descriptive Patterns and Their Application to Characterizing Classification Errors : %G eng %U http://hdl.handle.net/21.11116/0000-000C-165A-7 %U https://proceedings.mlr.press/v162/hedderich22a.html %D 2022 %B 39th International Conference on Machine Learning %Z date of event: 2022-07-17 - 2022-07-23 %C Baltimore, MA, USA %B Proceedings of the 39th International Conference on Machine Learning %E Chaudhuri, Kamalika; Jegelka, Stefanie; Le, Song; Csaba, Szepesvari; Gang, Niu; Sabato, Sivan %P 8691 - 8707 %B Proceedings of the Machine Learning Research %N 162 %@ false

%0 Conference Proceedings %A Ho, Vinh Thinh %A Stepanova, Daria %A Milchevski, Dragan %A Str&#246;tgen, Jannik %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations External Organizations External Organizations External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society %T Enhancing Knowledge Bases with Quantity Facts : %G eng %U http://hdl.handle.net/21.11116/0000-000A-614E-2 %R 10.1145/3485447.3511932 %D 2022 %B ACM Web Conference %Z date of event: 2022-04-25 - 2022-04-29 %C Virtual Event, Lyon, France %B WWW '22 %E Laforest, Fr&#233;d&#233;rique; Troncy, Rapha&#235;l; Simperl, Elena; Agarwal, Deepak; Gionis, Aristides; Herman, Ivan; M&#233;dini, Lionel %P 893 - 901 %I ACM %@ 978-1-4503-9096-5

%0 Thesis %A Ho, Vinh Thinh %Y Weikum, Gerhard %A referee: Stepanova, Daria %A referee: Theobald, Martin %+ Databases and Information Systems, MPI for Informatics, Max Planck Society International Max Planck Research School, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Entities with Quantities: Extraction, Search and Ranking : %G eng %U http://hdl.handle.net/21.11116/0000-000C-B756-5 %R 10.22028/D291-38030 %U urn:nbn:de:bsz:291--ds-380308 %F OTHER: hdl:20.500.11880/34538 %I Universit&#228;t des Saarlandes %C Saarbr&#252;cken %D 2022 %P xii, 131p. %V phd %9 phd %U https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/34538

%0 Conference Proceedings %E Kaffee, Lucie-Aim&#233;e %E Razniewski, Simon %E Amaral, Gabriel %E Alghamdi, Kholoud Saad %+ External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations External Organizations %T Wikidata Workshop 2022 : Proceedings of the 3rd Wikidata Workshop 2022, co-located with the 21st International Semantic Web Conference (ISWC2022) %G eng %U http://hdl.handle.net/21.11116/0000-000C-1663-C %U https://ceur-ws.org/Vol-3262/ %U urn:nbn:de:0074-3262-0 %I CEUR-WS %D 2022 %B 3rd Wikidata Workshop %Z date of event: - %C Virtual Event, Hangzhou, China %S CEUR Workshop Proceedings %V 3262

%0 Journal Article %A Lahoti, Preethi %A Gummadi, Krishna %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society %T Responsible Model Deployment via Model-agnostic Uncertainty Learning : %G eng %U http://hdl.handle.net/21.11116/0000-000B-58F0-3 %R 10.1007/s10994-022-06248-y %7 2022 %D 2022 %J Machine Learning %V 112 %& 939 %P 939 - 970 %I Springer %C Dordrecht %@ false

%0 Conference Proceedings %A Lahoti, Preethi %A Gummadi, Krishna %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society %T Detecting and Mitigating Test-time Failure Risks via Model-agnostic Uncertainty Learning : %G eng %U http://hdl.handle.net/21.11116/0000-000A-5E15-6 %R 10.1109/ICDM51629.2021.00141 %D 2022 %B 21st IEEE International Conference on Data Mining %Z date of event: 2021-12-07 - 2021-12-10 %C Auckland, New Zealand (Virtual Conference) %B 21st IEEE International Conference on Data Mining %E Bailey, James; Miettinen, Pauli; Koh, Yun Sing; Tao, Dacheng; Wu, Xindong %P 1174 - 1179 %I IEEE %@ 978-1-6654-2398-4

%0 Thesis %A Lahoti, Preethi %Y Weikum, Gerhard %A referee: Gummadi, Krishna %+ Databases and Information Systems, MPI for Informatics, Max Planck Society International Max Planck Research School, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Group K. Gummadi, Max Planck Institute for Software Systems, Max Planck Society %T Operationalizing Fairness for Responsible Machine Learning : %G eng %U http://hdl.handle.net/21.11116/0000-000A-CEC6-F %R 10.22028/D291-36586 %U nbn:de:bsz:291--ds-365860 %F OTHER: hdl:20.500.11880/33465 %I Universit&#228;t des Saarlandes %C Saarbr&#252;cken %D 2022 %P 129 p. %V phd %9 phd %X As machine learning (ML) is increasingly used for decision making in scenarios that impact humans, there is a growing awareness of its potential for unfairness. A large body of recent work has focused on proposing formal notions of fairness in ML, as well as approaches to mitigate unfairness. However, there is a growing disconnect between the ML fairness literature and the needs to operationalize fairness in practice. This thesis addresses the need for responsible ML by developing new models and methods to address challenges in operationalizing fairness in practice. Specifically, it makes the following contributions. First, we tackle a key assumption in the group fairness literature that sensitive demographic attributes such as race and gender are known upfront, and can be readily used in model training to mitigate unfairness. In practice, factors like privacy and regulation often prohibit ML models from collecting or using protected attributes in decision making. To address this challenge we introduce the novel notion of computationally-identifiable errors and propose Adversarially Reweighted Learning (ARL), an optimization method that seeks to improve the worst-case performance over unobserved groups, without requiring access to the protected attributes in the dataset. Second, we argue that while group fairness notions are a desirable fairness criterion, they are fundamentally limited as they reduce fairness to an average statistic over pre-identified protected groups. In practice, automated decisions are made at an individual level, and can adversely impact individual people irrespective of the group statistic. We advance the paradigm of individual fairness by proposing iFair (individually fair representations), an optimization approach for learning a low dimensional latent representation of the data with two goals: to encode the data as well as possible, while removing any information about protected attributes in the transformed representation. Third, we advance the individual fairness paradigm, which requires that similar individuals receive similar outcomes. However, similarity metrics computed over observed feature space can be brittle, and inherently limited in their ability to accurately capture similarity between individuals. To address this, we introduce a novel notion of fairness graphs, wherein pairs of individuals can be identified as deemed similar with respect to the ML objective. We cast the problem of individual fairness into graph embedding, and propose PFR (pairwise fair representations), a method to learn a unified pairwise fair representation of the data. Fourth, we tackle the challenge that production data after model deployment is constantly evolving. As a consequence, in spite of the best efforts in training a fair model, ML systems can be prone to failure risks due to a variety of unforeseen reasons. To ensure responsible model deployment, potential failure risks need to be predicted, and mitigation actions need to be devised, for example, deferring to a human expert when uncertain or collecting additional data to address model&#8217;s blind-spots. We propose Risk Advisor, a model-agnostic meta-learner to predict potential failure risks and to give guidance on the sources of uncertainty inducing the risks, by leveraging information theoretic notions of aleatoric and epistemic uncertainty. This dissertation brings ML fairness closer to real-world applications by developing methods that address key practical challenges. Extensive experiments on a variety of real-world and synthetic datasets show that our proposed methods are viable in practice. %U https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/33465

%0 Book %A Lin, Jimmy %A Nogueira, Rodrigo %A Yates, Andrew %+ External Organizations External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society %T Pretrained Transformers for Text Ranking : BERT and Beyond : %G eng %U http://hdl.handle.net/21.11116/0000-000C-DC2B-D %@ 978-3-031-02181-7 %R 10.1007/978-3-031-02181-7 %I Springer International Publishin %C Cham %D 2022 %P XVII, 307 %B Synthesis Lectures on Human Language Technologies

%0 Conference Proceedings %A Marx, Alexander %A Fischer, Jonas %+ External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society %T Estimating Mutual Information via Geodesic kNN : %G eng %U http://hdl.handle.net/21.11116/0000-0009-B19D-E %R 10.1137/1.9781611977172.47 %D 2022 %B SIAM International Conference on Data Mining %Z date of event: 2022-04-28 - 2022-04-30 %C Alexandria, VA, USA %B Proceedings of the SIAM International Conference on Data Mining %P 415 - 423 %I SIAM %@ 978-1-61197-717-2

%0 Report %A Nguyen, Thong %A Yates, Andrew %A Zirikly, Ayah %A Desmet, Bart %A Cohan, Arman %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations External Organizations External Organizations %T Improving the Generalizability of Depression Detection by Leveraging Clinical Questionnaires : %G eng %U http://hdl.handle.net/21.11116/0000-000C-166D-2 %U https://arxiv.org/abs/2204.10432 %D 2022 %X Automated methods have been widely used to identify and analyze mental health<br>conditions (e.g., depression) from various sources of information, including<br>social media. Yet, deployment of such models in real-world healthcare<br>applications faces challenges including poor out-of-domain generalization and<br>lack of trust in black box models. In this work, we propose approaches for<br>depression detection that are constrained to different degrees by the presence<br>of symptoms described in PHQ9, a questionnaire used by clinicians in the<br>depression screening process. In dataset-transfer experiments on three social<br>media datasets, we find that grounding the model in PHQ9's symptoms<br>substantially improves its ability to generalize to out-of-distribution data<br>compared to a standard BERT-based approach. Furthermore, this approach can<br>still perform competitively on in-domain data. These results and our<br>qualitative analyses suggest that grounding model predictions in<br>clinically-relevant symptoms can improve generalizability while producing a<br>model that is easier to inspect.<br> %K Computer Science, Computation and Language, cs.CL

%0 Conference Proceedings %A Nguyen, Thong %A Yates, Andrew %A Zirikly, Ayah %A Desmet, Bart %A Cohan, Arman %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations External Organizations External Organizations %T Improving the Generalizability of Depression Detection by Leveraging Clinical Questionnaires : %G eng %U http://hdl.handle.net/21.11116/0000-000B-1DAA-6 %R 10.18653/v1/2022.acl-long.578 %D 2022 %B 60th Annual Meeting of the Association for Computational Linguistic %Z date of event: 2022-05-22 - 2022-05-27 %C Dublin, Ireland %B The 60th Annual Meeting of the Association for Computational Linguistics %E Muresan, Smaranda; Nakov, Preslav; Villavicencio, Aline %P 8446 - 8459 %I ACL %@ 978-1-955917-21-6

%0 Report %A Nguyen, Tuan-Phong %A Razniewski, Simon %A Varde, Aparna %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society %T Extracting Cultural Commonsense Knowledge at Scale : %G eng %U http://hdl.handle.net/21.11116/0000-000B-58B3-8 %U https://arxiv.org/abs/2210.07763 %R 10.48550/arXiv.2210.07763 %D 2022 %X Structured knowledge is important for many AI applications. Commonsense<br>knowledge, which is crucial for robust human-centric AI, is covered by a small<br>number of structured knowledge projects. However, they lack knowledge about<br>human traits and behaviors conditioned on socio-cultural contexts, which is<br>crucial for situative AI. This paper presents CANDLE, an end-to-end methodology<br>for extracting high-quality cultural commonsense knowledge (CCSK) at scale.<br>CANDLE extracts CCSK assertions from a huge web corpus and organizes them into<br>coherent clusters, for 3 domains of subjects (geography, religion, occupation)<br>and several cultural facets (food, drinks, clothing, traditions, rituals,<br>behaviors). CANDLE includes judicious techniques for classification-based<br>filtering and scoring of interestingness. Experimental evaluations show the<br>superiority of the CANDLE CCSK collection over prior works, and an extrinsic<br>use case demonstrates the benefits of CCSK for the GPT-3 language model. Code<br>and data can be accessed at https://cultural-csk.herokuapp.com/.<br> %K Computer Science, Computation and Language, cs.CL,Computer Science, Artificial Intelligence, cs.AI

%0 Conference Proceedings %A Nguyen, Tuan-Phong %A Razniewski, Simon %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Materialized Knowledge Bases from Commonsense Transformers : %G eng %U http://hdl.handle.net/21.11116/0000-000B-1D87-D %U https://openreview.net/forum?id=HI5M4MYedZ5 %D 2022 %B 1st Workshop on Commonsense Representation and Reasoning %Z date of event: 2022-05-27 - 2022-05-27 %C Dublin, Ireland %B Proceedings of the First Workshop on Commonsense Representation and Reasoning %E Bosselut, Antoine; Li, Xiang; Yuchen, Bill; Shwartz, Vered; Majumder, Bodhisattwa Prasad; Kumar Lal, Yash; Rudinger, Rachel; Ren, Xiang; Tandon, Niket; Zouhar, Vil&#233;m %P 36 - 42 %I ACL %@ 978-1-955917-28-5

%0 Conference Proceedings %A Pradeep, Ronak %A Liu, Yuqi %A Zhang, Xinyu %A Li, Yilin %A Yates, Andrew %A Lin, Jimmy %+ External Organizations External Organizations External Organizations External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations %T Squeezing Water from a Stone: A Bag of Tricks for Further Improving Cross-Encoder Effectiveness for Reranking : %G eng %U http://hdl.handle.net/21.11116/0000-000A-9E28-8 %R 10.1007/978-3-030-99736-6_44 %D 2022 %B 44th European Conference on IR Research %Z date of event: 2022-04-10 - 2022-04-14 %C Stavanger, Norway %B Advances in Information Retrieval %E Hagen, Matthias; Verbene, Suzan; Macdonald, Craig; Seifert, Christin; Balog, Krisztian; N&#248;rv&#229;g, Kjetil; Setty, Vinay %P 655 - 670 %I Springer %@ 978-3-030-99736-6 %B Lecture Notes in Computer Science %N 13185

%0 Journal Article %A Puri, Manish %A Varde, Aparna S. %A de Melo, Gerard %+ External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations %T Commonsense Based Text Mining on Urban Policy : %G eng %U http://hdl.handle.net/21.11116/0000-000A-20AC-0 %R 10.1007/s10579-022-09584-6 %7 2022 %D 2022 %J Language Resources and Evaluation %O Computers and the Humanities Lang Resources & Evaluation %I Springer %C New York, NY %@ false %U https://rdcu.be/cJwGl

%0 Report %A Romero, Julien %A Razniewski, Simon %+ External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society %T Do Children Texts Hold The Key To Commonsense Knowledge? : %G eng %U http://hdl.handle.net/21.11116/0000-000B-58AA-3 %U https://arxiv.org/abs/2210.04530 %R 10.48550/arXiv.2210.04530 %D 2022 %X Compiling comprehensive repositories of commonsense knowledge is a<br>long-standing problem in AI. Many concerns revolve around the issue of<br>reporting bias, i.e., that frequency in text sources is not a good proxy for<br>relevance or truth. This paper explores whether children's texts hold the key<br>to commonsense knowledge compilation, based on the hypothesis that such content<br>makes fewer assumptions on the reader's knowledge, and therefore spells out<br>commonsense more explicitly. An analysis with several corpora shows that<br>children's texts indeed contain much more, and more typical commonsense<br>assertions. Moreover, experiments show that this advantage can be leveraged in<br>popular language-model-based commonsense knowledge extraction settings, where<br>task-unspecific fine-tuning on small amounts of children texts (childBERT)<br>already yields significant improvements. This provides a refreshing perspective<br>different from the common trend of deriving progress from ever larger models<br>and corpora.<br> %K Computer Science, Computation and Language, cs.CL,Computer Science, Artificial Intelligence, cs.AI

%0 Conference Proceedings %A Romero, Julien %A Razniewski, Simon %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Do Children Texts Hold The Key To Commonsense Knowledge? : %G eng %U http://hdl.handle.net/21.11116/0000-000C-DBE5-B %U https://aclanthology.org/2022.emnlp-main.752/ %D 2022 %B Conference on Empirical Methods in Natural Language Processing %Z date of event: 2022-12-07 - 2022-12-11 %C Abu Dhabi, United Arab Emirates %B Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing %E Goldberg, Yoav; Kozareva, Zornitsa; Zhang, Yue %P 10954 - 10959 %I ACL

%0 Conference Proceedings %E Singhania, Sneha %E Nguyen, Tuan-Phong %E Razniewski, Simon %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Knowledge Base Construction from Pre-trained Language Models 2022 : Proceedings of the Semantic Web Challenge on Knowledge Base Construction from Pre-trained Language Models 2022 co-located with the 21st International Semantic Web Conference (ISWC2022) %G eng %U http://hdl.handle.net/21.11116/0000-000B-C723-D %U urn:nbn:de:0074-3274-1 %U http://ceur-ws.org/Vol-3274/ %I CEUR-WS %D 2022 %B Semantic Web Challenge on Knowledge Base Construction from Pre-trained Language Models %Z date of event: 2022-10 - 2022-10 %D 2022 %C Virtual Event, Hanghzou, China %S CEUR Workshop Proceedings %V 3274

%0 Journal Article %A Singhania, Sneha %A Razniewski, Simon %A Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Predicting Document Coverage for Relation Extraction : %G eng %U http://hdl.handle.net/21.11116/0000-000A-27B8-B %R 10.1162/tacl_a_00456 %7 2022 %D 2022 %J Transactions of the Association of Computational Linguistics %V 10 %& 207 %P 207 - 223 %I ACL %C Cambridge, MA %@ false

%0 Thesis %A Tigunova, Anna %Y Weikum, Gerhard %A referee: Yates, Andrew %A referee: Demberg, Vera %+ Databases and Information Systems, MPI for Informatics, Max Planck Society International Max Planck Research School, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations %T Extracting Personal Information from Conversations : %G eng %U http://hdl.handle.net/21.11116/0000-000B-3FE1-1 %R 10.22028/D291-35628 %U nbn:de:bsz:291--ds-356280 %F OTHER: hdl:20.500.11880/32546 %I Universit&#228;t des Saarlandes %C Saarbr&#252;cken %D 2022 %P 139 p. %V phd %9 phd %X Personal knowledge is a versatile resource that is valuable for a wide range of downstream applications. Background facts about users can allow chatbot assistants to produce more topical and empathic replies. In the context of recommendation and retrieval models, personal facts can be used to customize the ranking results for individual users. A Personal Knowledge Base, populated with personal facts, such as demographic information, interests and interpersonal relationships, is a unique endpoint for storing and querying personal knowledge. Such knowledge bases are easily interpretable and can provide users with full control over their own personal knowledge, including revising stored facts and managing access by downstream services for personalization purposes. To alleviate users from extensive manual effort to build such personal knowledge base, we can leverage automated extraction methods applied to the textual content of the users, such as dialogue transcripts or social media posts. Mainstream extraction methods specialize on well-structured data, such as biographical texts or encyclopedic articles, which are rare for most people. In turn, conversational data is abundant but challenging to process and requires specialized methods for extraction of personal facts. In this dissertation we address the acquisition of personal knowledge from conversational data. We propose several novel deep learning models for inferring speakers&#8217; personal attributes: &#8226; Demographic attributes, age, gender, profession and family status, are inferred by HAMs - hierarchical neural classifiers with attention mechanism. Trained HAMs can be transferred between different types of conversational data and provide interpretable predictions. &#8226; Long-tailed personal attributes, hobby and profession, are predicted with CHARM - a zero-shot learning model, overcoming the lack of labeled training samples for rare attribute values. By linking conversational utterances to external sources, CHARM is able to predict attribute values which it never saw during training. &#8226; Interpersonal relationships are inferred with PRIDE - a hierarchical transformer-based model. To accurately predict fine-grained relationships, PRIDE leverages personal traits of the speakers and the style of conversational utterances. Experiments with various conversational texts, including Reddit discussions and movie scripts, demonstrate the viability of our methods and their superior performance compared to state-of-the-art baselines. %U https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/32546

%0 Conference Proceedings %A Tran, Hai Dang %A Yates, Andrew %+ Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations %T Dense Retrieval with Entity Views : %G eng %U http://hdl.handle.net/21.11116/0000-000C-1669-6 %R 10.1145/3511808.3557285 %D 2022 %B 31st ACM International Conference on Information and Knowledge Management %Z date of event: 2022-10-17 - 2022-10-21 %C Atlanta GA USA %B CIKM '22 %E Al Hasan, Mohammad; Xiong, Li %P 1955 - 1964 %I ACM %@ 978-1-4503-9236-5

%0 Journal Article %A Varde, Aparna S. %+ Databases and Information Systems, MPI for Informatics, Max Planck Society %T Computational Estimation by Scientific Data Mining with Classical Methods to Automate Learning Strategies of Scientists : %G eng %U http://hdl.handle.net/21.11116/0000-000A-9D92-0 %R 10.1145/3502736 %7 2022 %D 2022 %J ACM Transactions on Knowledge Discovery from Data %V 16 %N 5 %& 1 %P 1 - 52 %Z sequence number: 86 %I ACM %C New York, NY

%0 Journal Article %A Varde, Aparna S. %A Pandey, Abidha %A Du, Xu %+ Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations External Organizations %T Prediction Tool on Fine Particle Pollutants and Air Quality for Environmental Engineering : %G eng %U http://hdl.handle.net/21.11116/0000-000A-2F55-3 %R 10.1007/s42979-022-01068-2 %7 2022 %D 2022 %J SN Computer Science %V 3 %N 3 %Z sequence number: 184 %I Springer Nature %C Singapore %@ false

%0 Thesis %A Wang, Yongqing %Y Pal, Koninika %A referee: Weikum, Gerhard %+ Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society %T Coreference Resolution for Extracting Quantity-Facts from Multiple Sentences : %G eng %U http://hdl.handle.net/21.11116/0000-000C-F5F1-F %I Universit&#228;t des Saarlandes %C Saarbr&#252;cken %D 2022 %P XI, 58 p. %V master %9 master %X Commonsense knowledge (CSK) about concepts and their properties is useful for AI applications such as robust chatbots. Prior works like ConceptNet, TupleKB and others compiled large CSK collections, but are restricted in their expressiveness to subject-predicate-object (SPO) triples with simple concepts for S and monolithic strings for P and O. Also, these projects have either prioritized precision or recall, but hardly reconcile these complementary goals. This thesis presents a methodology, called Ascent, to automatically build a large-scale knowledge base (KB) of CSK assertions, with advanced expressiveness and both better precision and recall than prior works. Ascent goes beyond triples by capturing composite concepts with subgroups and aspects, and by refining assertions with semantic facets. The latter are important to express temporal and spatial validity of assertions and further qualifiers. Ascent combines open information extraction with judicious cleaning using language models. Intrinsic evaluation shows the superior size and quality of the Ascent KB, and an extrinsic evaluation for QA-support tasks underlines the benefits of Ascent.