Seminar: Ausgewählte Themen des Web 2.0

Veranstalter: Prof. Dr. Andreas Hotho
Florian Lautenschlager, Thomas Niebler, Albin Zehe.

Bei Fragen, Reservierung eines Themas, … wenden Sie sich bitte an Florian Lautenschlager.

WueCampus-Kurs: zum Kurs (Zugang wird nach der Vorbesprechung individuell freigeschaltet)

BibSonomy Gruppe: uw_ss17_web2.0

Vorbesprechung: 26.04.2017 14:00 Uhr im Raum B002

Themenvergabe und Vorbesprechung

Themen können schon vor dem Vorbesprechungstermin reserviert werden. Die Themen werden einige Zeit vor der Vorbesprechung hier auf dieser Seite bekannt gegeben.

Bei der Vorbesprechung werden die Themen - so nicht schon vergeben - an die anwesenden Seminarteilnehmer verteilt. Grundsätzlich besteht Anwesenheitspflicht. Teilnehmer, die bereits ein Thema bearbeiten, aber ohne Absprache fehlen, verlieren den Anspruch auf ihr Thema. In der Vorbesprechung werden auch die Termine für die Vorträge festgelegt.

Konzept und Ablauf des Seminars

Im Zentrum dieses Seminars steht ein wissenschaftlicher Artikel und ein darin verwendetes Konzept. Ein Beispiel wäre das Paper

HypTrails: A Bayesian Approach for Comparing Hypotheses About Human Trails on the Web
Singer, Philipp and Helic, Denis and Hotho, Andreas and Strohmaier, Markus

In dem folgendes Konzept eine zentrale Rolle spielt:

Bayes Factor

Im Rahmen des Seminars soll nun das Konzept aufgearbeitet und verwandte Literatur gesucht werden. Das Ziel ist es, in Ausarbeitung und dem Vortrag des Seminars das gegebene Konzept unter Verwendung entsprechender Referenzen verständlich einzuführen und zu erklären. Außerdem sollen Anwendungsbeispiele oder Erweiterungen des Konzepts aus drei anderen Artikeln beleuchtet werden (dies kann den bereits gegebenen Artikel mit einschliessen).

Der Ablauf des Seminar ist wie folgt:

Zuerst wird die Ausarbeitung angefertigt.
Dann wird die Arbeit abgegeben und begutachtet. Diese Begutachtung erfolgt durch den Betreuer und anonym durch zwei andere Seminarteilnehmer. Das heisst, jeder Seminarteilnehmer, muss je zwei Beurteilungen für andere Seminarteilnehmer anfertigen. Die Begutachtungen gehen zurück an den Author, der dann die Möglichkeit hat, seine Ausarbeitung zu überarbeiten.
Dann wird zeitgleich der Vortrag (20 Minuten Vortrag, 5 Minuten Fragen) und eine einseitige Zusammenfassung vorbereitet.
Der Vortrag wird dann in Blockveranstaltungen gehalten. Die oben genannten Materialien (Ausarbeitung, Vortrag, Kurzzusammenfassung) müssen vor der ersten Blockveranstaltung abgegeben werden. Bei den Blackveranstaltungen besteht Anwesenheitspflicht.

Themen

Die Grundlage zu den Vorträgen bilden wissenschaftliche Aufsätze (Papers). Jedes hier aufgeführte Paper entspricht einem Thema.

themenrelevante Paper sind mit dem Tag thema versehen
die einzelnen Themen werden anhand der Tags thema:x, thema:y, etc. unterschieden
Ziel des Seminars ist es NICHT den Inhalt des Papers zusammenzufassen, sondern das Thema, das durch das Tag thema:x gegeben ist, aufzuarbeiten.
die Tags findet man auf BibSonomy in der entsprechenden Gruppe: uw_ss17_web2.0

Die Themen werden in den nächsten Wochen bekannt gegeben.

A Study of Synthetic Oversampling for Twitter Imbalanced Sentiment Analysis. Ah-Pine, Julien; Soriano-Morales, Edmundo-Pavel. In DMNLP@PKDD/ECML, Bd. 1646 von CEUR Workshop Proceedings, P. Cellier, T. Charnois, A. Hotho, S. Matwin, M.-F. Moens, Y. Toussaint (Hrsg.), S. 17–24. CEUR-WS.org, 2016.
- [ BibTeX ]
- [ URL ]
@inproceedings{conf/pkdd/Ah-PineS16, author = {Ah-Pine, Julien and Soriano-Morales, Edmundo-Pavel}, booktitle = {DMNLP@PKDD/ECML}, crossref = {conf/pkdd/2016dmnlp}, editor = {Cellier, Peggy and Charnois, Thierry and Hotho, Andreas and Matwin, Stan and Moens, Marie-Francine and Toussaint, Yannick}, keywords = {oversampling}, pages = {17-24}, publisher = {CEUR-WS.org}, series = {CEUR Workshop Proceedings}, title = {A Study of Synthetic Oversampling for Twitter Imbalanced Sentiment Analysis.}, volume = 1646, year = 2016 }
Adversarial examples in the physical world. Kurakin, Alexey; Goodfellow, Ian; Bengio, Samy. 2016.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Most existing machine learning classifiers are highly vulnerable to adversarial examples. An adversarial example is a sample of input data which has been modified very slightly in a way that is intended to cause a machine learning classifier to misclassify it. In many cases, these modifications can be so subtle that a human observer does not even notice the modification at all, yet the classifier still makes a mistake. Adversarial examples pose security concerns because they could be used to perform an attack on machine learning systems, even if the adversary has no access to the underlying model. Up to now, all previous work have assumed a threat model in which the adversary can feed data directly into the machine learning classifier. This is not always the case for systems operating in the physical world, for example those which are using signals from cameras and other sensors as an input. This paper shows that even in such physical world scenarios, machine learning systems are vulnerable to adversarial examples. We demonstrate this by feeding adversarial images obtained from cell-phone camera to an ImageNet Inception classifier and measuring the classification accuracy of the system. We find that a large fraction of adversarial examples are classified incorrectly even when perceived through the camera.

@misc{kurakin2016adversarial, abstract = {Most existing machine learning classifiers are highly vulnerable to adversarial examples. An adversarial example is a sample of input data which has been modified very slightly in a way that is intended to cause a machine learning classifier to misclassify it. In many cases, these modifications can be so subtle that a human observer does not even notice the modification at all, yet the classifier still makes a mistake. Adversarial examples pose security concerns because they could be used to perform an attack on machine learning systems, even if the adversary has no access to the underlying model. Up to now, all previous work have assumed a threat model in which the adversary can feed data directly into the machine learning classifier. This is not always the case for systems operating in the physical world, for example those which are using signals from cameras and other sensors as an input. This paper shows that even in such physical world scenarios, machine learning systems are vulnerable to adversarial examples. We demonstrate this by feeding adversarial images obtained from cell-phone camera to an ImageNet Inception classifier and measuring the classification accuracy of the system. We find that a large fraction of adversarial examples are classified incorrectly even when perceived through the camera.}, author = {Kurakin, Alexey and Goodfellow, Ian and Bengio, Samy}, keywords = {thema:adversarial-examples}, note = {cite arxiv:1607.02533Comment: 14 pages, 6 figures. Demo available at https://youtu.be/zQ_uMenoBCk}, title = {Adversarial examples in the physical world}, year = 2016 }
A critique of word similarity as a method for evaluating distributional semantic models. Batchkarov, Miroslav; Kober, Thomas; Reffin, Jeremy; Weeds, Julie; Weir, David. In ACL 2016, S. 7. 2016.
- [ BibTeX ]
@article{batchkarov2016critique, author = {Batchkarov, Miroslav and Kober, Thomas and Reffin, Jeremy and Weeds, Julie and Weir, David}, journal = {ACL 2016}, keywords = {semantics}, pages = 7, title = {A critique of word similarity as a method for evaluating distributional semantic models}, year = 2016 }
Multi-Scale Convolutional Neural Networks for Time Series Classification. Cui, Zhicheng; Chen, Wenlin; Chen, Yixin. 2016.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Time series classification (TSC), the problem of predicting class labels of time series, has been around for decades within the community of data mining and machine learning, and found many important applications such as biomedical engineering and clinical prediction. However, it still remains challenging and falls short of classification accuracy and efficiency. Traditional approaches typically involve extracting discriminative features from the original time series using dynamic time warping (DTW) or shapelet transformation, based on which an off-the-shelf classifier can be applied. These methods are ad-hoc and separate the feature extraction part with the classification part, which limits their accuracy performance. Plus, most existing methods fail to take into account the fact that time series often have features at different time scales. To address these problems, we propose a novel end-to-end neural network model, Multi-Scale Convolutional Neural Networks (MCNN), which incorporates feature extraction and classification in a single framework. Leveraging a novel multi-branch layer and learnable convolutional layers, MCNN automatically extracts features at different scales and frequencies, leading to superior feature representation. MCNN is also computationally efficient, as it naturally leverages GPU computing. We conduct comprehensive empirical evaluation with various existing methods on a large number of benchmark datasets, and show that MCNN advances the state-of-the-art by achieving superior accuracy performance than other leading methods.

@misc{cui2016multiscale, abstract = {Time series classification (TSC), the problem of predicting class labels of time series, has been around for decades within the community of data mining and machine learning, and found many important applications such as biomedical engineering and clinical prediction. However, it still remains challenging and falls short of classification accuracy and efficiency. Traditional approaches typically involve extracting discriminative features from the original time series using dynamic time warping (DTW) or shapelet transformation, based on which an off-the-shelf classifier can be applied. These methods are ad-hoc and separate the feature extraction part with the classification part, which limits their accuracy performance. Plus, most existing methods fail to take into account the fact that time series often have features at different time scales. To address these problems, we propose a novel end-to-end neural network model, Multi-Scale Convolutional Neural Networks (MCNN), which incorporates feature extraction and classification in a single framework. Leveraging a novel multi-branch layer and learnable convolutional layers, MCNN automatically extracts features at different scales and frequencies, leading to superior feature representation. MCNN is also computationally efficient, as it naturally leverages GPU computing. We conduct comprehensive empirical evaluation with various existing methods on a large number of benchmark datasets, and show that MCNN advances the state-of-the-art by achieving superior accuracy performance than other leading methods.}, author = {Cui, Zhicheng and Chen, Wenlin and Chen, Yixin}, keywords = {thema}, note = {cite arxiv:1603.06995}, title = {Multi-Scale Convolutional Neural Networks for Time Series Classification}, year = 2016 }
Fast R-CNN. Girshick, Ross. In CoRR, abs/1504.08083. 2015.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper proposes a Fast Region-based Convolutional Network method (Fast R-CNN) for object detection. Fast R-CNN builds on previous work to efficiently classify object proposals using deep convolutional networks. Compared to previous work, Fast R-CNN employs several innovations to improve training and testing speed while also increasing detection accuracy. Fast R-CNN trains the very deep VGG16 network 9× faster than R-CNN, is 213× faster at test-time, and achieves a higher mAP on PASCAL VOC 2012. Compared to SPPnet, Fast R-CNN trains VGG16 3× faster, tests 10× faster, and is more accurate. Fast R-CNN is implemented in Python and C++ (using Caffe) and is available under the open-source MIT License at https: //github.com/rbgirshick/fast-rcnn.

@article{noauthororeditor, abstract = {This paper proposes a Fast Region-based Convolutional Network method (Fast R-CNN) for object detection. Fast R-CNN builds on previous work to efficiently classify object proposals using deep convolutional networks. Compared to previous work, Fast R-CNN employs several innovations to improve training and testing speed while also increasing detection accuracy. Fast R-CNN trains the very deep VGG16 network 9× faster than R-CNN, is 213× faster at test-time, and achieves a higher mAP on PASCAL VOC 2012. Compared to SPPnet, Fast R-CNN trains VGG16 3× faster, tests 10× faster, and is more accurate. Fast R-CNN is implemented in Python and C++ (using Caffe) and is available under the open-source MIT License at https: //github.com/rbgirshick/fast-rcnn.}, author = {Girshick, Ross}, journal = {CoRR}, keywords = {object}, title = {Fast R-CNN}, volume = {abs/1504.08083}, year = 2015 }
LINE: Large-scale Information Network Embedding. Tang, Jian; Qu, Meng; Wang, Mingzhe; Zhang, Ming; Yan, Jun; Mei, Qiaozhu. In WWW, S. 1067–1077. ACM, Florence, Italy, 2015.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper studies the problem of embedding very large information networks into low-dimensional vector spaces, which is useful in many tasks such as visualization, node classification, and link prediction. Most existing graph embedding methods do not scale for real world information networks which usually contain millions of nodes. In this paper, we propose a novel network embedding method called the ``LINE,'' which is suitable for arbitrary types of information networks: undirected, directed, and/or weighted. The method optimizes a carefully designed objective function that preserves both the local and global network structures. An edge-sampling algorithm is proposed that addresses the limitation of the classical stochastic gradient descent and improves both the effectiveness and the efficiency of the inference. Empirical experiments prove the effectiveness of the LINE on a variety of real-world information networks, including language networks, social networks, and citation networks. The algorithm is very efficient, which is able to learn the embedding of a network with millions of vertices and billions of edges in a few hours on a typical single machine. The source code of the LINE is available online\footnote{\url{https://github.com/tangjianpku/LINE}}.

@inproceedings{tang2015largescale, abstract = {This paper studies the problem of embedding very large information networks into low-dimensional vector spaces, which is useful in many tasks such as visualization, node classification, and link prediction. Most existing graph embedding methods do not scale for real world information networks which usually contain millions of nodes. In this paper, we propose a novel network embedding method called the ``LINE,'' which is suitable for arbitrary types of information networks: undirected, directed, and/or weighted. The method optimizes a carefully designed objective function that preserves both the local and global network structures. An edge-sampling algorithm is proposed that addresses the limitation of the classical stochastic gradient descent and improves both the effectiveness and the efficiency of the inference. Empirical experiments prove the effectiveness of the LINE on a variety of real-world information networks, including language networks, social networks, and citation networks. The algorithm is very efficient, which is able to learn the embedding of a network with millions of vertices and billions of edges in a few hours on a typical single machine. The source code of the LINE is available online\footnote{\url{https://github.com/tangjianpku/LINE}}.}, address = {New York, NY, USA}, author = {Tang, Jian and Qu, Meng and Wang, Mingzhe and Zhang, Ming and Yan, Jun and Mei, Qiaozhu}, booktitle = {WWW}, keywords = {deepwiki}, pages = {1067–1077}, publisher = {ACM}, title = {LINE: Large-scale Information Network Embedding}, year = 2015 }
Bilingual word representations with monolingual quality in mind. Luong, Thang; Pham, Hieu; Manning, Christopher D. In Proceedings of the 1st Workshop on Vector Space Modeling for Natural Language Processing, S. 151–159. 2015.
- [ BibTeX ]
@inproceedings{luong2015bilingual, author = {Luong, Thang and Pham, Hieu and Manning, Christopher D}, booktitle = {Proceedings of the 1st Workshop on Vector Space Modeling for Natural Language Processing}, keywords = {word2vec}, pages = {151–159}, title = {Bilingual word representations with monolingual quality in mind}, year = 2015 }
DRAW: A Recurrent Neural Network For Image Generation. Gregor, Karol; Danihelka, Ivo; Graves, Alex; Rezende, Danilo Jimenez; Wierstra, Daan. 2015.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper introduces the Deep Recurrent Attentive Writer (DRAW) neural network architecture for image generation. DRAW networks combine a novel spatial attention mechanism that mimics the foveation of the human eye, with a sequential variational auto-encoding framework that allows for the iterative construction of complex images. The system substantially improves on the state of the art for generative models on MNIST, and, when trained on the Street View House Numbers dataset, it generates images that cannot be distinguished from real data with the naked eye.

@misc{gregor2015recurrent, abstract = {This paper introduces the Deep Recurrent Attentive Writer (DRAW) neural network architecture for image generation. DRAW networks combine a novel spatial attention mechanism that mimics the foveation of the human eye, with a sequential variational auto-encoding framework that allows for the iterative construction of complex images. The system substantially improves on the state of the art for generative models on MNIST, and, when trained on the Street View House Numbers dataset, it generates images that cannot be distinguished from real data with the naked eye.}, author = {Gregor, Karol and Danihelka, Ivo and Graves, Alex and Rezende, Danilo Jimenez and Wierstra, Daan}, keywords = {generation}, note = {cite arxiv:1502.04623}, title = {DRAW: A Recurrent Neural Network For Image Generation}, year = 2015 }
Development of NO2 and NOx land use regression models for estimating air pollution exposure in 36 study areas in Europe – The ESCAPE project. Beelen, Rob; Hoek, Gerard; Vienneau, Danielle; Eeftens, Marloes; Dimakopoulou, Konstantina; Pedeli, Xanthi; Tsai, Ming-Yi; Künzli, Nino; Schikowski, Tamara; Marcon, Alessandro; Eriksen, Kirsten T.; Raaschou-Nielsen, Ole; Stephanou, Euripides; Patelarou, Evridiki; Lanki, Timo; Yli-Tuomi, Tarja; Declercq, Christophe; Falq, Grégoire; Stempfelet, Morgane; Birk, Matthias; Cyrys, Josef; von Klot, Stephanie; Nádor, Gizella; Varró, Mihály János; Dėdelė, Audrius; Gražulevičienė, Regina; Mölter, Anna; Lindley, Sarah; Madsen, Christian; Cesaroni, Giulia; Ranzi, Andrea; Badaloni, Chiara; Hoffmann, Barbara; Nonnemacher, Michael; Krämer, Ursula; Kuhlbusch, Thomas; Cirach, Marta; de Nazelle, Audrey; Nieuwenhuijsen, Mark; Bellander, Tom; Korek, Michal; Olsson, David; Strömgren, Magnus; Dons, Evi; Jerrett, Michael; Fischer, Paul; Wang, Meng; Brunekreef, Bert; de Hoogh, Kees. In Atmospheric Environment, 72, S. 10–23. 2013.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Estimating within-city variability in air pollution concentrations is important. Land use regression (LUR) models are able to explain such small-scale within-city variations. Transparency in LUR model development methods is important to facilitate comparison of methods between different studies. We therefore developed LUR models in a standardized way in 36 study areas in Europe for the ESCAPE (European Study of Cohorts for Air Pollution Effects) project. Nitrogen dioxide (NO2) and nitrogen oxides (NOx) were measured with Ogawa passive samplers at 40 or 80 sites in each of the 36 study areas. The spatial variation in each area was explained by LUR modelling. Centrally and locally available Geographic Information System (GIS) variables were used as potential predictors. A leave-one out cross-validation procedure was used to evaluate the model performance. There was substantial contrast in annual average NO2 and NOx concentrations within the study areas. The model explained variances (R2) of the LUR models ranged from 55% to 92% (median 82%) for NO2 and from 49% to 91% (median 78%) for NOx. For most areas the cross-validation R2 was less than 10% lower than the model R2. Small-scale traffic and population/household density were the most common predictors. The magnitude of the explained variance depended on the contrast in measured concentrations as well as availability of GIS predictors, especially traffic intensity data were important. In an additional evaluation, models in which local traffic intensity was not offered had 10% lower R2 compared to models in the same areas in which these variables were offered. Within the ESCAPE project it was possible to develop LUR models that explained a large fraction of the spatial variance in measured annual average NO2 and NOx concentrations. These LUR models are being used to estimate outdoor concentrations at the home addresses of participants in over 30 cohort studies.

@article{BEELEN201310, abstract = {Estimating within-city variability in air pollution concentrations is important. Land use regression (LUR) models are able to explain such small-scale within-city variations. Transparency in LUR model development methods is important to facilitate comparison of methods between different studies. We therefore developed LUR models in a standardized way in 36 study areas in Europe for the ESCAPE (European Study of Cohorts for Air Pollution Effects) project. Nitrogen dioxide (NO2) and nitrogen oxides (NOx) were measured with Ogawa passive samplers at 40 or 80 sites in each of the 36 study areas. The spatial variation in each area was explained by LUR modelling. Centrally and locally available Geographic Information System (GIS) variables were used as potential predictors. A leave-one out cross-validation procedure was used to evaluate the model performance. There was substantial contrast in annual average NO2 and NOx concentrations within the study areas. The model explained variances (R2) of the LUR models ranged from 55% to 92% (median 82%) for NO2 and from 49% to 91% (median 78%) for NOx. For most areas the cross-validation R2 was less than 10% lower than the model R2. Small-scale traffic and population/household density were the most common predictors. The magnitude of the explained variance depended on the contrast in measured concentrations as well as availability of GIS predictors, especially traffic intensity data were important. In an additional evaluation, models in which local traffic intensity was not offered had 10% lower R2 compared to models in the same areas in which these variables were offered. Within the ESCAPE project it was possible to develop LUR models that explained a large fraction of the spatial variance in measured annual average NO2 and NOx concentrations. These LUR models are being used to estimate outdoor concentrations at the home addresses of participants in over 30 cohort studies.}, author = {Beelen, Rob and Hoek, Gerard and Vienneau, Danielle and Eeftens, Marloes and Dimakopoulou, Konstantina and Pedeli, Xanthi and Tsai, Ming-Yi and Künzli, Nino and Schikowski, Tamara and Marcon, Alessandro and Eriksen, Kirsten T. and Raaschou-Nielsen, Ole and Stephanou, Euripides and Patelarou, Evridiki and Lanki, Timo and Yli-Tuomi, Tarja and Declercq, Christophe and Falq, Grégoire and Stempfelet, Morgane and Birk, Matthias and Cyrys, Josef and von Klot, Stephanie and Nádor, Gizella and Varró, Mihály János and Dėdelė, Audrius and Gražulevičienė, Regina and Mölter, Anna and Lindley, Sarah and Madsen, Christian and Cesaroni, Giulia and Ranzi, Andrea and Badaloni, Chiara and Hoffmann, Barbara and Nonnemacher, Michael and Krämer, Ursula and Kuhlbusch, Thomas and Cirach, Marta and de Nazelle, Audrey and Nieuwenhuijsen, Mark and Bellander, Tom and Korek, Michal and Olsson, David and Strömgren, Magnus and Dons, Evi and Jerrett, Michael and Fischer, Paul and Wang, Meng and Brunekreef, Bert and de Hoogh, Kees}, journal = {Atmospheric Environment}, keywords = {pollution}, pages = {10 - 23}, title = {Development of NO2 and NOx land use regression models for estimating air pollution exposure in 36 study areas in Europe – The ESCAPE project}, volume = 72, year = 2013 }
Semantics made by you and me: Self-emerging ontologies can capture the diversity of shared knowledge. Benz, Dominik; Hotho, Andreas; Stützer, Stefan; Stumme, Gerd. In WebSci. 2010.
- [ BibTeX ]
@inproceedings{benz2010semantics, author = {Benz, Dominik and Hotho, Andreas and Stützer, Stefan and Stumme, Gerd}, booktitle = {WebSci}, keywords = {thema:ontology_learning}, title = {Semantics made by you and me: Self-emerging ontologies can capture the diversity of shared knowledge}, year = 2010 }
Taking sides: user classification for informal online political discourse. Malouf, Robert; Mullen, Tony. In Internet Research, 18, S. 177–190. 2008.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Purpose – To evaluate and extend, existing natural language processing techniques into the domain of informal online political discussions. Design/methodology/approach – A database of postings from a US political discussion site was collected, along with self‐reported political orientation data for the users. A variety of sentiment analysis, text classification, and social network analysis methods were applied to the postings and evaluated against the users' self‐descriptions. Findings – Purely text‐based methods performed poorly, but could be improved using techniques which took into account the users' position in the online community. Research limitations/implications – The techniques we applied here are fairly simple, and more sophisticated learning algorithms may yield better results for text‐based classification. Practical implications – This work suggests that social network analysis is an important tool for performing natural language processing tasks with informal web texts. Originality/value – This research extends sentiment analysis to a new subject domain (US politics) and a new text genre (informal online discusssions).

@article{noauthororeditor, abstract = {Purpose – To evaluate and extend, existing natural language processing techniques into the domain of informal online political discussions. Design/methodology/approach – A database of postings from a US political discussion site was collected, along with self‐reported political orientation data for the users. A variety of sentiment analysis, text classification, and social network analysis methods were applied to the postings and evaluated against the users' self‐descriptions. Findings – Purely text‐based methods performed poorly, but could be improved using techniques which took into account the users' position in the online community. Research limitations/implications – The techniques we applied here are fairly simple, and more sophisticated learning algorithms may yield better results for text‐based classification. Practical implications – This work suggests that social network analysis is an important tool for performing natural language processing tasks with informal web texts. Originality/value – This research extends sentiment analysis to a new subject domain (US politics) and a new text genre (informal online discusssions).}, author = {Malouf, Robert and Mullen, Tony}, journal = {Internet Research}, keywords = {user}, pages = {177-190}, title = {Taking sides: user classification for informal online political discourse}, volume = 18, year = 2008 }
Learning a distance metric from relative comparisons. Schultz, Matthew; Joachims, Thorsten. In NIPS. 2004.
- [ BibTeX ]
- [ URL ]
@inproceedings{schultz2004learning, author = {Schultz, Matthew and Joachims, Thorsten}, booktitle = {NIPS}, keywords = {metric}, title = {Learning a distance metric from relative comparisons}, year = 2004 }

Leistungsnachweis

Die Ausarbeitung muss dem Betreuer zum bekanntgegebenen Termin zum Begutachten als PDF vorliegen.

Die zwei Begutachtungen anderer Ausarbeitungen, die müssen pünktlich zum bekanntgegebenen Termin angefertigt und abgegeben werden.

Ein (1!) Tag vor dem ersten geplanten Vortrag sind per Email als PDF dem Betreuer abzugeben:

eine 6-seitige LaTeX-Ausarbeitung im hier beschriebenen Format (doppelspaltig)
eine einseitige Zusammenfassung (die allen Teilnehmern beim Vortrag ausgehändigt wird),
der Foliensatz

Vorherige Absprachen mit dem Betreuer sind ausdrücklich erwünscht. Alle verwendeten Referenzen sind zusätzlich zum Literaturverzeichnis der Ausarbeitung in http://www.bibsonomy.org in der entsprechenden Gruppe (relevant for) mit den folgenden Tags einzugeben:

den Tags des jeweiligen Themas und weiteren sinnvollen Tags.

Die Vortragsdauer beträgt verbindlich 20 Minuten, nach dem Vortrag besteht Gelegenheit zur Diskussion (ca. 5 Minuten). Der Vortrag (inkl. Folien und Abstract) geht mit 30% in die Endnote ein, die Ausarbeitung ebenfalls mit 30%. Die angefertigeten Reviews gehen 20% in die Note ein. Die letzten 20% bewerten die Eigenständigkeit der Ausarbeitung, Einhaltung von Terminen, Beteiligung am Seminar etc.

Bearbeitungshinweise

Als Richtlinie für die Erstellung einer guten Seminararbeit (inkl. Vortrag und Ausarbeitung) wird das Buch

Markus Deininger and Horst Lichter and Jochen Ludewig and Kurt Schneider. Studien-Arbeiten: ein Leitfaden zur Vorbereitung, Durchführung und Betreuung von Studien-, Diplom- Abschluss- und Doktorarbeiten am Beispiel Informatik. 5. Auflage. vdf Hochschulverlag, Zürich, 2005.

empfohlen, welches beim Betreuer des Seminars (Prof. Dr. Andreas Hotho) ausliegt. Wir empfehlen die Anschaffung dieses Buchs (9,50 €), da es Sie bis zur Masterarbeit (und weiter) begleiten kann. Die Benotung der Seminararbeit erfolgt in Anlehnung an das dort auf Seite 77 angegebene Schema, angepasst auf die Erfordernisse einer Seminararbeit.

Seminar: Ausgewählte Themen des Web 2.0