Supplementary Proceedings of the Sixth International Conference on Analysis of Images, Social Networks and Texts (AIST-SUP 2017), Moscow, Russia, July 27-29, 2017
AIST is a scientific conference on Analysis of Images, Social Networks, and Texts. The conference is intended for computer scientists and practitioners whose research interests involve Internet mathematics and other related fields of data science. Similar to the previous year, the conference will be focused on applications of data mining and machine learning techniques to various problem domains: image processing, analysis of social networks, and natural language processing. We hope that the participants will benefit from the interdisciplinary nature of the conference and exchange experience.
The paper proposes a list of requirements for a game able to describe individually motivated social interactions: be non-cooperative, able to construct multiple coalitions in an equilibrium and incorporate intra and inter coalition externalities. For this purpose the paper presents a family of non-cooperative games for coalition structure construction with an equilibrium existence theorem for a game in the family. Few examples illustrate the approach. One of the results is that efficiency is not equivalent to cooperation as an allocation in one coalition. Further papers will demonstrate other applications of the approach.
We consider deep reinforcement learning algorithms for playing a game based on video input. We compare choosing proper hyper-parameters in deep Q-network model and model-free episodic control focused on reusing of successful strategies. The evaluation was made based on Pong video game implemented in Unreal Engine 4.
Commonly in network analysis a graph (network) is represented by its adjacency matrix, and the latter may have an enormous order. We show that in many situations (generalizing the case of regular graph) a much smaller matrix (referred as type adjacency matrix) may be used instead. We introduce concepts of the types of nodes and of the type adjacency matrix, study properties of the latter and demonstrate some of its applications in social and economic network analysis. In particular, we consider centrality measures in undirected networks and dynamic patterns in a development model based on the structure of optimal paths in directed weighted networks.
The article describes the ontology-based approach to systematization and search of academic English style markers. The designed ontology is divided into two levels: the first level provides the information about linguistic terms and the second consists of style markers, which were derived by experts in linguistic. It is suggested to generate lexical-semantic template based on the ontology to identify the list of markers in the text with the help of Domain Specific Language (DSL) technology. Currently, there is JAPE-template (Java Annotation Patterns Engine) of GATE text processing system.
The paper presents the result of the research of the inuence of text font size on attention indicators. On the basis of the experimental data, the multiple linear regression of the dependence of the optimum of the font size on the criterion of maximizing the value of mental efficiency indicator from the indicators of attention and memory of the subject was constructed. An algorithm for adapting the font size of text for optimal perception is presented.
In this paper, we present an object-attribute grammar (OAG) – an original formalism for describing the natural language semantic analysis algorithm for the linguistic processor (LP). Special focus is made on formatting input and output data for LP. The LP uses the graph containing word interpretations of source text and transforms it to the graph representing meaning of the text. We present definition of graph transformation algorithm of LP by means of the developed notation, denoted a template of a subgraph which is searched in a graph and an operation of subgraph transformation. The software implementation of the LP based on OAG is described.
This paper presents a method to analyze discussions from social network by using deep learning. We have prepared a new dataset by collecting discussions from a social network and annotating remarks of the discussion. The annotation consists of two types of labels for each message: intention type and direction of intention. Using this dataset and pre-trained word embeddings we have evaluated two neural network structures. On the basis of evaluation, we chose a model to automatically predict intention types and direction of intention of an arbitrary message from any social network
This research examines the problems of automatic scientific articles classification according to Universal Decimal Classifier. To reveal the structure of the train data its visualization was obtained using the recursive feature elimination algorithm. Further; the study provides a comparison of TF-IDF and Weirdness – two statistic-based metrics of keyword significance. The most efficient classification methods are explained: cosine similarity method, naïve Bayesian classifier and artificial neural network. This research explores the most effective for text categorization structure of the multi-layer perceptron and derives appropriate conclusions.
We propose using NB-SVM over bag of character n-grams input representation for determining part-of-speech tags and grammatical categories like gender, number, etc. for words in Russian texts. Several methods are compared including CRF (Conditional Random Fields), SVM (Support Vector Machines) and NB-SVM (Naive Bayes SVM) and superiority of NB-SVM over other classifiers is shown. The proposed model is the 5th best among 12 other models in the first shared task of the MorphoRuEval-17 challenge. We also experimented with category grouping when a single classifier is used to determine several grammatical categories and showed that it improves the model per- formance even further.