Analysis of Images, Social Networks and Texts. 6th International Conference, 2017, Revised Selected Papers
This volume contains the refereed proceedings of the 6th International Conference on Analysis of Images, Social Networks, and Texts (AIST 2017)1. The previous conferences during 2012–2016 attracted a significant number of students, researchers, academics, and engineers working on interdisciplinary data analysis of images, texts, and social networks. The broad scope of AIST made it an event where researchers from different domains, such as image and text processing, exploiting various data analysis techniques, can meet and exchange ideas. We strongly believe that this may lead to cross fertilisation of ideas between researchers relying on modern data analysis machinery. Therefore, AIST brought together all kinds of applications of data mining and machine learning techniques. The conference allowed specialists from different fields to meet each other, present their work, and discuss both theoretical and practical aspects of their data analysis problems. Another important aim of the conference was to stimulate scientists and people from industry to benefit from the knowledge exchange and identify possible grounds for fruitful collaboration. The conference was held during July 27–29, 2017. The conference was organised in Moscow, the capital of Russia, on the campus of Moscow Polytechnic University. This year, the key topics of AIST were grouped into six tracks: 1. General topics of data analysis chaired by Sergei Kuznetsov (Higher School of Economics, Russia) and Amedeo Napoli (LORIA, France) 2. Natural language processing chaired by Natalia Loukachevitch (Lomonosov Moscow State University, Russia) and Alexander Panchenko (University of Hamburg, Germany) 3. Social network analysis chaired by Stanley Wasserman (Indiana University, USA) 4. Analysis of images and video chaired by Victor Lempitsky (Skolkovo Institute of Science and Technology, Russia) and Andrey Savchenko (Higher School of Economics, Russia) 5. Optimisation problems on graphs and network structures chaired by Panos Pardalos (University of Florida, USA) and Michael Khachay (IMM UB RAS and Ural Federal University, Russia) 6. Analysis of dynamic behaviour through event data chaired by Wil van der Aalst (Eindhoven University of Technology, The Netherlands) and Irina Lomazova (Higher School of Economics, Russia) One of the novelties this year was the introduction of a new specialised track on process mining (Track 6).
The paper deals with Google’s universal parser SyntaxNet. The system was used to analyze the Universal Dependencies linguistic corpora. We conducted an error analysis of the output of the parser to reveal to what extent the error types are connected with or preconditioned by the language types. In particular, we carried out several experiments, clustering the languages based on the frequency of different errors made by SyntaxNet, and studied the similarity of the resulting clustering with the traditional typology of languages. Three types of errors were separately considered: part-of-speech tagging, dependency labeling, and attachment errors. We show that there is indeed a correlation between error frequencies and language types, which might indicate that to further improve the performance of a universal parser, one needs to take into account language-specific morphological and syntactic structures.
Homophily - tendency for people to form social connections with similar others - is one of the key topics in social network analysis. It indicates to what extent people tend to be similar to their friends and in what dimensions. For the long time homophily was just an index of the social similarity, but for the recent years the interest for the homophily formation, dynamics and multidimensionality increased. In this paper we investigate the homophily in such social constructed behavior as food consumption and academic achievements. The study of body mass index in social network context reveals the presence of homophily, which means that persons with similar constitution are more likely to be interconnected with each other. Interestingly, that healthy food consumption has no impact on social network formation, but there is homophily based on fast food consumption. Thus, ‘bad habits’ are stronger forces for the social ties formation. This results show that social constructed behavior is an important component on the process of social network formation.
In this paper we present an approach for searching sub-traces in event logs, generated by information systems. Our technique is heavily based on the Aho-Corasick algorithm, and extends it with simultaneous search on several event log traces. The computational complexity of the proposed approach was estimated. Moreover, the approach was implemented and verified on real-life event logs. It was shown that it allows to reduce the search time for event logs with a high proportion of similar traces.
Consider a continuous word embedding model. Usually, the cosines between word vectors are used as a measure of similarity of words. These cosines do not change under orthogonal transformations of the embedding space. We demonstrate that, using some canonical orthogonal transformations from SVD, it is possible both to increase the meaning of some components and to make the components more stable under re-learning. We study the interpretability of components for publicly available models for the Russian language (RusVectores, fastText, RDT).
In this paper we propose the two-stage approach of organizing information in video surveillance systems. At first, the faces are detected in each frame and a video stream is split into sequences of frames with face region of one person. Secondly, these sequences (tracks) that contain identical faces are grouped using face verification algorithms and hierarchical agglomerative clustering. Gender and age are estimated for each cluster (person) in order to facilitate the usage of the organized video collection. The particular attention is focused on the aggregation of features extracted from each frame with the deep convolutional neural networks. The experimental results of the proposed approach using YTF and IJB-A datasets demonstrated that the most accurate and fast solution is achieved for matching of normalized average of feature vectors of all frames in a track.
Modern co-authorship networks contain hidden patterns of researchers interaction and publishing activities. We aim to provide a system for selecting a collaborator for joint research or an expert on a given list of topics. We have improved a recommender system for finding possible collaborator with respect to research interests and predicting quality and quantity of the anticipated publications. Our system is based on a co-authorship network derived from the bibliographic database, as well as content information on research papers obtained from SJR Scimago, staff information and the other features from the open data of researchers profiles. We formulate the recommendation problem as a weighted link prediction within the co-authorship network and evaluate its prediction for strong and weak ties in collaborative communities.
The natural language structure can be viewed as weighted semantic network. Such representation gives an option to investigate the text corpus as the model of the subject domain. In this paper we propose the mechanism of the semantic network identification and construction. We apply the methodological instrument for the social media text analysis and trace the dynamics of the discussions about 1917 year within the internet communities. Network changes illustrate the changes of the interest to different topics. The proposed mechanism can be used for the monitoring of the different social processes and phenomenal in online social networks and media.
This paper presents a human gait data collection for analysis and activity recognition consisting of continues recordings of combined activities, such as walking, running, taking stairs up and down, sitting down, and so on; and the data recorded are segmented and annotated. Data were collected from a body sensor network consisting of six wearable inertial sensors (accelerometer and gyroscope) located on the right and left thighs, shins, and feet. Additionally, two electromyography sensors were used on the quadriceps (front thigh) to measure muscle activity. This database can be used not only for activity recognition but also for studying how activities are performed and how the parts of the legs move relative to each other. Therefore, the data can be used (a) to perform health-care-related studies, such as in walking rehabilitation or Parkinson’s disease recognition, (b) in virtual reality and gaming for
In this paper, we present novel winning team predicting models and compare the accuracy of the obtained prediction with TrueSkill model of ranking individual players impact based on their impact in team victory for the two most popular online games: Dota 2 and Counter-Strike: Global Offensive.
The paper presents an open-source morphological processor of Russian texts recently developed and named CrossMorphy. The processor performs lemmatization, morphological tagging of both dictionary and non-dictionary words, contextual and non-contextual morphological disambiguation, generation of word forms, as well as morphemic parsing of words. Besides the extended functionality, emphasis is put on linguistic quality of word processing and easy integration into programming projects. CrossMorphy is fully implemented in C++ programming language on the base of OpenCorpora vocabulary data.
In this research we analyze the demand for performing arts. Since the observed demand is limited by the capacity of house, one needs to account for demand censorship. The presence of consumer segments with different purposes of going to the theatre and willingness-to-pay for performance and ticket characteristics compels to account for heterogeneity in theatre demand. In this paper we propose an estimator for prediction of demand that accounts for both demand censorship and preferences heterogeneity. The estimator is based on the idea of classification and regression trees and bagging prediction aggregation. We extend the algorithm for censored data prediction problem. Our algorithm predicts and combines predictions from both discrete and continuous parts of censored data. We show that the estimator is better in prediction accuracy compared with estimators which account for censorship or heterogeneity of preferences only.
This paper deals with automatic classification of questions in the Russian language, a natural early step in building a question answering system. We developed a typology of Russian questions using interrogative particles, pronouns and word order as the main features. A corpus of 2008 questions was manually compiled and annotated according to our typology. We used a fine-grained class set and a coarse-grained one (23 and 14 classes, respectively). The training data, represented as character bi-/trigrams and word uni-/bi-/trigrams, was used to approach the task of question classification. We tested several widely used machine-learning methods (logistic regression, support vector machines, naïve Bayes) against a regular expression baseline on a held-out test corpus annotated by an external expert. The best results were achieved by a SVM classifier (linear kernel) that achieved the accuracy of 65.3% (fine-grained) and 68.7% (coarse-grained), while the baseline regular expression model showed 52.7% accuracy.
Vision-based simultaneous localization and mapping (vSLAM) is a well-established problem in mobile robotics and monocular vSLAM is one of the most challenging variations of that problem nowadays. In this work we study one of the core post-processing optimization mechanisms in vSLAM, e.g. loop-closure detection. We analyze the existing methods and propose original algorithm for loop-closure detection, which is suitable for dense, semi-dense and feature-based vSLAM methods. We evaluate the algorithm experimentally and show that it contribute to more accurate mapping while speeding up the monocular vSLAM pipeline to the extent the latter can be used in real-time for controlling small multi-rotor vehicle (drone).