NewsREEL Multimedia at MediaEval’18

Author: Benjamin Kille

This year’s edition of our news recommendation challenge NewsREEL focused on multimedia data. We asked participants to estimate which articles would become popular solely based on their textual and visual features. A large-scale data set collected by our long-term partners at plista facilitated evaluating different algorithms.

The MediaEval benchmark brings different evaluation tasks together. This year’s edition took place in the time from 29 to 31 October in Nice, France. The event offers task organisers the opportunity to present their challenges. Participants can discuss ideas and illustrate their results.

Three papers have been submitted for NewsREEL Multimedia. The overview paper [1] outlines the task, details the evaluation methodology, and presents the results. The baseline paper [2] illustrates the baselines against which we compared participants’ predictions. Ciobanu et al. [3] analysed how Google’s Vision API can be used to obtain more representative labels for images.

Besides the technical programme, the MediaEval organisers had prepared social events. On the first day, we visited the Château-Musée Grimaldi and avoided drowning during the subsequent city tour. On the second day, we enjoyed dinner at Chez David. Finally, on the third day, we dined at Château Le Cagnard. During the diners, we had the opportunity to discuss and exchange ideas with a diverse group of researchers. We have come up with ideas to foster future cooperation.

Next year, MediaEval will return to Nice colocated with ACM Multimedia.


[1] Lommatzsch, A., Kille, B., Hopfgartner, F. and Ramming, L., 2018, October. NewsREEL Multimedia at MediaEval 2018: News Recommendation with Image and Text Content. In Working Notes Proceedings of the MediaEval 2018 Workshop. CEUR-WS.


[2] Lommatzsch, A. and Kille, B., 2018. Baseline Algorithms for Predicting the Interest in News based on Multimedia Data.In Working Notes Proceedings of the MediaEval 2018 Workshop. CEUR-WS.


[3] Ciobanu, A., Lommatzsch, A. and Kille, B., 2018. Predicting the Interest in News based On Image Annotations. In Working Notes Proceedings of the MediaEval 2018 Workshop. CEUR-WS.


CIKM and INRA’19 in Turin

Author: Benjamin Kille

Turin hosted the 27th edition of Conference on Information and Knowledge Management (CIKM) on 22-26 October 2018. The first day featured nine workshops. We co-organised the sixth edition of the International Workshop for News Recommendation and Analytics (INRA). Between twenty and thirty attendees listened to three keynote talks. Frank Hopfgartner, senior lecturer at Sheffield University, presented a comprehensive review of information retrieval as well as recommender systems evaluation initiatives. Anja Benner-Tischler, legal scholar at Kassel University, outlined the recently introduced EU General Data Protection Regulations (GDPR). Leif Ramming, the lead of plista‘s machine learning team, discussed issues related to scaling up recommender systems on an industrial level. In addition, attendees followed six paper presentations. The seventh paper could not be presented in person due to visa issues.

The second day commenced with the keynote speech by Maarten de Riijke. He highlighted the interactive aspects of environments, in which todays information access systems operate. Subsequently, the conference split into three rounds with five parallel sessions each. The day concluded with the first of three short, demo, and industry sessions. The third day kicked off with Edward Grefenstette‘s keynote about how modelling rewards affects agents’ learning of language. The remainder of the day followed the same structure as before with three rounds of five parallel sessions followed by the second short, demo, and industry session. Yoelle Maarek started the fourth day with her keynote about Amazon’s Alexa. Subsequently, attendees split up to listen to the final rounds of research talks. Besides the final short, demo, and industry session, the programme included a townhall discussion. The final day offered a selection of tutorials.

In 2019, CIKM will take place in Beijing, China on 3-7 November.

The TU Berlin participates in the Festival of Lights 2018

Author: Andreas Lommatzsch

Every year in October the Festival of Light Berlin illuminates over 50 buildings in the city. In 2018 the TU Berlin participated in the festival for the first time. Microscopic images showing structures of leafs and crystals are projected on the TU-Tower at the Ernst-Reuter-Platz (one of the tallest buildings in the Western Part of Berlin). Impressions from our office and the Ernst-Reuter-Platz are shown in the following photos.

KI 2018 in Berlin

Author: Andreas Lommatzsch

The 41th edition of the German AI conference ("KI 2018") has been held in Berlin, September 24-28, 2018. Starting with two days with workshops and tutorials, the main conference ran from Wednesday to Friday. Each day of the main conference started with an interesting keynote. A highlight was the keynote given by Dietmar Jannach on session-based recommendation approaches. Professor Jannach discussed the transition from rating prediction to ranking with implicit feedback. He stressed that recommender systems research has to critically reflect on the desired output rather than marginally improve arbitrary criteria.

CC IRML contributed two-fold the conference:
(1) On Monday, we held a half-day tutorial on stream-based recommendation algorithms. The tutorial discussed how to extend existing recommender algorithms and evaluation to stream-based scenarios. In addition, we explained the NewsREEL challenge, which offers the possibility of evaluating stream-based recommender algorithms both online and offline. About twenty participants attended and engaged in discussions.
(2) On Friday, B. Jain presented his work on "Condorcet’s Jury Theorem for Consensus Clustering." The talk discussed the theoretical justification for the consensus clustering method.

In 2019 the KI conference will be held in Kassel (23-26 Sept 2019); in 2020 the conference will come to Bamberg.

LWDA Conference in Mannheim

Authors: Andreas Lommatzsch and Asmaa Haja

The conference "Lernen Wissen, Daten, Analysen" has been held in Darmstadt August 22-24, 2018. The conference is organized by the German Computer Science Society (GI) and consists of 5 tracks: Information Retrieval (FGIR 2018), Knowledge Management (FGWM 2018), Knowledge Discovery, Data Mining, and Machine Learning (KDML 2018, Business Intelligence (WSBI 2018), and Large-Scale Data Management and Processing – Applications in Research and Industry (FGDB 2018).

The conferences program consists of research paper presentation, a poster session and four interesting keynotes from different domains. The first keynote was given by Daniela Nicklas. The talk discussed the question of who watches the sensors, which are used for monitoring real-world phenomena. The second keynote, given by Kerstin Bach, explained the "selfBACK" decision support system (DSS) project. The system uses Case-Based Reasoning for supporting patients with low back pains. Frank Giesler presented the third talk. He introduced a tried-and-tested approach that ensures successful data-driven digital product development through the close involvement of future users. The last talk, given by Stephan Mandt, gave an overview of some exciting recent developments in deep probabilistic modeling, which combines deep neural networks with probabilistic models for unsupervised learning.

The DAI-Labor presented the research conducted in the student project. Our paper "Normalization of Time series for Improving Recommendations" studies how observed log data in recommender systems can be decomposed into several different characteristic patterns. We explain that the decomposition is the basis for understanding the lifecycle of the items and the characteristic user preferences. Furthermore, the decomposition can be used for trend-based prediction, extrapolating the extracted time series taking into account the characteristic patterns. The methods have been studied in a News Recommendation scenario that is characterized by a continuously changing set of items and strong variances in the number of users during the day.
In addition to a talk, we also presented a poster allowing us to discuss details of the scenario and our approach individually.

Overall, the conference provided many interesting talks covering a wide spectrum of topics. The positive atmosphere has been the basis for exciting discussion.

EYZ Media Research: Cooperation With DAI-Labor TU Berlin

Authors: Jing Juan, Andreas Lommatzsch, Antje Marx

Recommender systems have proven effective to help the audience of streaming platforms finding relevant movies and serials hidden in huge amount of available content. The question is how can the audience be assisted in finding the most interesting movies matching the individual user preferences. Popular platforms, like Netflix, Pandora, and Amazon successfully provide recommender systems. Collaborative filtering based approach (item-based and user-based) are popular to provide recommendations.

Standard recommendation approaches have several weaknesses in niche market due to the user data sparsity, missing ratings, and highly specific items. Therefore, customized recommendation strategies are needed, adapting to the specific system features while catering users’ demand. And such customization can take place when traditional recommender algorithms tailored to the special items format and user experience in the corresponding niche markets, thereby add value to the specific system or platform.
This has been the motivation for the cooperation between EYZ Media and the DAI-Labor of the TU Berlin.

The collaboration between EYZ Media and DAI-Lab in year 2018 considers both traditional recommendation strategies (content-based approach and collaborative filtering approach) and modern event-based approach (introducing domain relevant trend in heterogeneous system resource like Twitter). Obeying the API and dataflow definition from EYZ Media, DAI-Labor provides the support on recommendation solution by building up elasticsearch service, collaborative filtering component and event-based recommender pipeline. Corresponding evaluation turned out from both experts’ subjective opinion and offline evaluation on specific metrics help us better understand how recommenders booster user activities in the niche market.

Project details are discussed in the realeyz tech blog.


The DAI-Labor (Distributed Artificial Intelligence Laboratory) at the TU Berlin conducts research and development projects in order to provide solutions for a new generation of artificial intelligence systems and services. The competence center Information Retrieval and Machine Learning (IRML) focuses on developing innovative recommender systems and information retrieval systems combining machine learning techniques, graph-based approaches and social media analysis.


EYZ Media operates the VOD platform realeyz, on the web at, as Prime Video Channels and as an app in the stores. Thanks to its focus on independent film, realeyz has built up a growing user base and is one of the top 3 VOD specialty providers in Germany. RealEYZ stands for intelligent, exciting, classic and innovative content in all formats and lengths, curated carefully and with passion. Every day a new film, all in its original version with a selection of 1,000+ titles – and the focus on indie movies appeals above all to young, metropolitan users. The use of many unique components created by EYZ’s own dev. Team such as the recommender system with its algorithms that enable individual and personal UX as well as realeyz’’s presence on different channels and technical environments sharpen the profile, strengthen customer loyalty and establish the realeyz brand as the essential hub for independent productions. RealEYZ is supported by the EU-CREATIVE program and is part of the EuroVoD network.

Supported by Investitonsbank Berlin

1st Annual Meeting of the German Chinese Association for Artificial Intelligence

Author: Andreas Lommatzsch

The first annual meeting of the German Chinese Association for Artificial Intelligence (GCAAI) took place at Porsche Digital Lab, on July 27th, 2018.

GCAAI promotes the exchange of educational resources between Germany and Chine in the field of artificial intelligence. They organize meetups and seminars focusing on AI topics. The events feature speakers from academia and industry. Participants gain access to valuable information, have the opportunity to discuss recent trends with scholars, and receive support to find internships or job opportunities.

The meetup at the Porsche Digital Lab has been an exciting event for discussing current projects and research activities in Artificial Intelligence. The meeting started with six short presentations: Dr. Han Xiao, a Senior Research Scientist at Tencent-AI, presented the challenges when teaching machines to comprehend texts and to answer questions. He gave an excellent overview of methods for extracting and summarizing information. In addition, the talk discussed the problem of detecting questions unrelated to a given text/domain.

In the meetup, I gave a short introduction about recommender system and presented the NewsREEL challenge as well as the MediaEval NewsREEL Benchmark. I highlighted the research questions and the unique chance to evaluate news recommender algorithms with consideration of multi-media data coming from an operating news recommender system.

The talks have been a good starting point for fruitful discussions. The meeting continued with a dinner in the Indochinese restaurant in Kreuzberg. The next meeting of the GCAAI is planned for November.

Announcing the 6th International Workshop on News Recommendation and Analytics (INRA 2018)

Author: Andreas Lommatzsch

The 27th edition of the Conference on Knowledge Management and Machine Learning will take place in Torino (Italy) October 22-26. In 2018, the conference focuses on big data, big information, and big knowledge. Thus, the topics of interest cover a wide range which includes data acquisition, pre-processing, analysis, visualization, evaluation, and innovative applications.

The news domain generates a plethora of data leading to various problems in the realm of "big data". We want to address these issues and more in the sixth edition of the "Workshop on News Recommendation and Analytics (INRA)" in conjunction with CIKM.

Continuously changing streams of unstructured, fragmentary stories necessitate research on how to categorize, verify, and present news. A multitude of challenging research questions emerges as a result. The workshop brings together researchers, media companies, and practitioners to exchange ideas and explore ways to support creating, distributing, and recommending digital news.

The primary topics of interests are:

  • News Recommendation
  • News Analytics
  • Fake News and Disinformation
  • User Experience Issues
  • Evaluation Platforms, Methods and Datasets

We also provide Data Sets and Evaluation Platforms if you would like to test their ideas on real world news settings.

The Keynote Talks will be given by

  • Frank Hopfgartner, Information School of University of Sheffield, UK
  • Anja Benner-Tischler, University of Kassel, Germany
  • Leif Ramming, Plista GmbH, Germany

Use the EasyChair Conference system to submit contributions. Accepted papers will be published in CEUR workshop series by CIKM.

Please do not forget the important dates:

INRA 2018

6th International Workshop on News Recommendation and Analytics (INRA 2018)

Presenting our Chatbot Framework at the Conference of Innovative Internet Community Services

Author: Andreas Lommatzsch

The 18th edition of the International Conference on Innovative Internet Community Systems (I4CS) took place at the University of Zilina, Slovakia, 18-20 June 2018.
This year the conference focuses was on innovation methods for enterprises, methods for building the Internet of Things (IoT) as well as innovative applications.

I presented our chatbot framework optimized for answering questions related to public administrations’ services. In contrast to most existing frameworks, our approach provides detailed answers to several hundred services and helps users in a dialog to find all information relevant for a specific information need. The system combines several data sources (databases, geo information, ontologies, and synonyms) and uses several different natural language processing methods for detecting the user’s intension. In my talk, I presented our approach in detail and discussed our experiences operating the chatbot system live on the service web portal of the Berlin’s administration. Find more details in the conference paper.

The conference featured a multitude of exciting presentations. I would like to highlight the talk “Innovation Management Methods in the Aviation Industry” given by Karl-Heinz Lüke. The presentation discussed the results of a survey of external and in-house innovation methods giving profound insights into the aviation industry’s way of thinking.

The conference’s main social event was a hiking tour in the Mala Fatra with its magnificent gorges and beautiful hills. The tour has fostered discussing trends and experiences concerning current research projects in an inspiring landscape.

In 2019, the 19th edition of the I4CS conference will be held in Wolfsburg, Germany.

NewsREEL Multimedia @ MediaEval Benchmark 2018

Author: Andreas Lommatzsch

The MediaEval Benchmarking is an Initiative for Multimedia Evaluation offering data and challenges in the form of shared tasks. MediaEval focusses on innovative algorithms and methods for exploring and retrieval multi-media data. The provided data typically cover several modalities such as audio-visual (video, audio), textual, and contextual data.

NewsREEL Multimedia (News Recommendation with Image/Text Contents) runs as a new pilot task in 2018. The task is based on the observation that images have a big impact on perception of news. The NewsREEL MultiMedia provides a data collection of news articles, item popularity data, and (compressed) image data. The task consists in predicting the popularity of new items based on the multi-media data. The quality of the predictions is measured with the Precision and the NDCG.

The MediaEval workshop 2018 will be held at EURECOM in Sophia Antipolis, France, in November 2018.

Web page: MediaEval-Benchmark

Web page: MediaEval-Benchmark