Tutorials
Diachronic word embeddings for semantic shifts modeling: how to trace changes of meaning in time
Andrey Kutuzov, University of Oslo
Biography
Andrey Kutuzov is currently a doctoral research fellow at the University of Oslo, Norway. He holds a Masters degree in Computational Linguistics from the National Research University Higher School of Economics (Moscow, Russia). His industrial experience includes working with Lionbridge and Mail.ru Search.

Andrey has published papers at the top-tier international conference like EMNLP, ACL, CONLL, EACL and COLING. His primary research interest is using distributional word embedding models to trace long-term and short-term semantic shifts (see the survey paper on this).

Andrey is the leading maintainer of the RusVectōrēs project featuring various word embedding models for Russian and their visualizations. He teaches courses in NLP and deep learning at Masters level.
to be announced
Paper Presentations
Comparative Analysis of Scientific Papers Collections via Topic Modeling and Co-authorship Networks
Fedor Krasnov, Alexander Dimentov and Mikhail Shvartsman
Experimental Comparison of Unsupervised Approaches in the Task of Separating Specializations within Professions in Job Vacancies
Mikhail Vinel, Ivan Ryazanov, Dmitry Botov and Ivan Nikolaev
Effects of Training Data Size and Class Imbalance on the Performance of Classifiers
Wanwan Zheng and Mingzhe Jin
Soft Estimates of User Protection from Social Engineering Attacks: Fuzzy Combination of User Vulnerabilities and Malefactor Competencies in the Attacking Impact Success Prediction
Maxim Abramov and Alexander Tulupyev
Retrieval of Visually Shared News
Dmitrijs Milajevs
Usage of HMM-based Speech Recognition Methods for Automated Determination of a Similarity Level between Languages
Ansis Ataols Bērziņš
Binary Autoencoder for Text Modeling
Ruslan Baynazarov and Irina Piontkovskaya
Bi-LSTM Model for Morpheme Segmentation of Russian Words
Elena Bolshakova and Alexander Sapin
Prosodic Boundaries Prediction in Russian Using Morphological and Syntactic Features
Alla Menshikova and Daniil Kocharov
SentiRusColl: Russian Collocation Lexicon for Sentiment Analysis
Anastasia Kotelnikova and Evgeny Kotelnikov
An Approach to Abstractive Summarization for Norwegian Bokmål
Mariia Fedorova and Valentin Malykh
An Approach to Inter Annotator Agreement Evaluation for the Named Entity Recognition Task
Liliya Volkova and Viktor Bocharov
poster and demo session
to be announced
industrial session
Increasing the value and impact of NLP-based solutions. An overview of SILO.AI use cases and learnings from them.
Luiza Sayfullina, Silo.AI
Biography

Luiza Sayfullina is a senior Machine learning expert with 7+ years of Machine Learning experience having a deep understanding of Natural Language Processing for English and Finnish language. Proven track record from helping companies find and implement AI solutions ranging from low resource text classification, information extraction, summarization to speech-to-text applications. Organizer of ML study groups in Helsinki since 2016. PhD in Neural networks and Natural Language Processing from Aalto University (2019). Luiza's special interest lies also in the intersection of psychology and AI.
Language Technology for Financial Markets
Dmitry Kan, AlphaSense Inc.
Biography

Dmitry Kan is the Head of Search with AlphaSense, premier business insights platform. Lately he has been focusing on Research projects. In his daily work he builds search algorithms, researching and using advances in Natural Language Processing and machine learning. In this talk Dmitry will explain the unique challenges for Language Technology as applied to financial domain. We will take a look at how our algorithms power the world's most fastest moving knowledge workers in a wide range of industries. Dmitry holds PhD in Computer Science from Saint Petersburg State University. His interests include search engines, sentiment analysis, recommender systems and language modelling.
Natural Language Processing With (Almost) No Language Resources
Amir Bakarov, Huawei
Long story short: imagine you have a regular NLP task (classification, sequence labeling, ranking, etc). If you have enough data, you can employ BERT (or a more fancy Transformer-based LM) and fine-tune it and feel good. Or not, if you have only around a dozen training samples. But is it ever possible to achieve good performance of a supervised model with a very few amount of data?

In this talk I will overview the approaches to dealing with the problem of learning on small amount of data other than the traditional "just hire more assessors". I am going to survey some recent advances in semi-supervised learning, transfer learning and weak supervision, and tell how these techniques could be applied to ubiqutious NLP tasks.

Biography

Amir Bakarov is currently a research engineer at Huawei Technologies (St. Petersburg Research Centre). He holds a Masters degree in Computational Linguistics from the National Research University Higher School of Economics (Moscow, Russia). His experience includes working both in academia and industry. His primary research interests are distributional semantics and methods of analysis and interpretation of meaning representations based on distributional hypothesis.
to be announced
Feel free to contact us at ainlevent@gmail.com