Analysis of Images, Social Networks and Texts. 8th International Conference, AIST 2019, Lecture Notes in Computer Science, Revised Selected Papers
This book constitutes the post-conference proceedings of the 8th International Conference on Analysis of Images, Social Networks and Texts, AIST 2019, held in Kazan, Russia, in July 2019.
The 27 full and 8 short papers were carefully reviewed and selected from 134 submissions (of which 21 papers were automatically rejected without being reviewed). The papers are organized in topical sections on general topics of data analysis; natural language processing; social network analysis; analysis of images and video; optimization problems on graphs and network structures; and analysis of dynamic behavior through event data.
In this paper we focus on the problem of multi-label image recognition for visually-aware recommender systems. We propose a two stage approach in which a deep convolutional neural network is firstly fine-tuned on a part of the training set. Secondly, an attention-based aggregation network is trained to compute the weighted average of visual features in an input image set. Our approach is implemented as a mobile fashion recommender system application. It is experimentally show on the Amazon Fashion dataset that our approach achieves an F1-measure of 0.58 for 15 recommendations, which is twice as good as the 0.25 F1-measure for conventional averaging of feature vectors.
Double-blind peer reviewing has been proved to be a pretty effective and fair way of academic work selection. However, to the best of our knowledge, nobody has yet analysed the effects caused by its introduction at the Russian NLP conferences. We investigate how the double-blind peer reviewing influences gender and location (according to authors’ affiliations) biases and whether it makes two of the conferences under analysis more inclusive. The results show that gender distribution has become more equal for the Dialogue conference, but did not change for the AIST conference. The authors’ location distribution (roughly divided into ‘central’ and ‘not central’) has become more equal for AIST, but, interestingly, less equal for Dialogue.
In this exploratory study, we analyze reading behavior using logs from an ebook reading app. The logs contain users’ page turns along with time stamps and page sizes in characters. We focus on 17 readers of War and Peace by Leo Tolstoy, who read at least 80% of the novel. We aim at learning a regression model for reading speed based on shallow textual (e.g. word and sentence lengths) and contextual (e.g. time of the day and position in the book) features. Contextual features outperform textual ones and allow to predict reading speed with moderate quality. We share insights about the results and outline directions for future research. The analysis of reading behavior can be beneficial for school education, reading promotion, book recommendation, as well as for traditional creative writing and interactive fiction design.
This work tackles the problem of modeling author style in Russian. In particular, we solve the task of authorship attribution using the collected dataset of 30 authors, 1506 texts written in the period of 18th – 21st century. We apply various approaches to solving the attribution problem: Random Forest, Logistic Regression, SVM Classifier. In terms of text representation, we use seven models in three language levels: lexis, morphology, and syntax. Most importantly, we propose our own set of morpho-syntactic features that perform on about the same level as doc2vec, but are fully interpretable. The conducted experiments show the effectiveness of their standalone use, as well as the increase in the quality of classification when using these attributes along with the classic doc2vec-based approach. All code, including feature extraction, is made freely available. Additionally, we analyze the performance of individual features as style markers. Finally, we study classification errors in order to identify the patterns in the misattribution of specific authors.
In this paper we improve the speed of the nearest neighbor classiﬁers of a set of points based on sequential analysis of high-dimensional feature vectors. Each input object is associated with a sequence of principal component scores of aggregated features extracted by deep neural network. The number of components in each element of this sequence is dynamically chosen based on explained proportion of total variance for the training set.We propose to process the next element with higher explained variance only if the decision for the current element is unreliable. This reliability is estimated by matching of the ratio of the minimum distance and all other distances with a certain threshold. Experimental study for face recognition with the Labeled Faces in the Wild and YouTube Faces datasets demonstrates the decrease of running time up to 10 times when compared to conventional instance-based learning.
This is an application of an advanced entity recognition algorithm to a large dataset.
Due to the development of the Internet and social networks, civil participation in political processes is taking new forms, not accounted for in the classical theoretical frameworks. In this study, we present a new methodology for researching these new, unconventional forms of civil participation. We use data collected from a social network site VK.com. Social network analysis methods, such as multilevel exponential random graph models (ERGM) were used to analyze the data. First proposed by Wang in 2013, multilevel ERGM allows us to model tie formation in a wide class of social networks. This method allows us to efficiently model group membership based on node attributes. Since the collected data is fundamentally 2-mode, this model allows us to identify the important factors that lead people to join online protest communities.