Analysis of Images, Social Networks and Texts. 7th International Conference AIST 2018
This book constitutes the proceedings of the 7th International Conference on Analysis of Images, Social Networks and Texts, AIST 2018, held in Moscow, Russia, in July 2018.
The 29 full papers were carefully reviewed and selected from 107 submissions (of which 26 papers were rejected without being reviewed). The papers are organized in topical sections on natural language processing; analysis of images and video; general topics of data analysis; analysis of dynamic behavior through event data; optimization problems on graphs and network structures; and innovative systems.
Process mining is a new discipline aimed at constructing process models from event logs. Recently several methods for the discovery of transition systems from event logs were introduced. Considering these transition systems as finite state machines classical algorithms for deriving regular expressions can be applied. Regular expressions allow representing sequential process models in a hierarchical way, using sequence, choice, and iterative patterns. The aim of this work is to apply and tune an algorithm deriving regular expressions from transition systems within the process mining domain.
This paper deals with automatic classification of questions in the Russian language. In contrast to previously used methods, we introduce a convolutional neural network for question classification. We took advantage of an existing corpus of 2008 questions, manually annotated in accordance with a pragmatic 14-class typology. We modified the data by reducing the typology to 13 classes, expanding the dataset and improving the representativeness of some of the question types. The training data in a combined representation of word embeddings and binary regular expression-based features was used for supervised learning to approach the task of question tagging. We tested a convolutional neural network against a state-of-the-art Russian language question classification algorithm, an SVM classifier with a linear kernel and questions represented as word trigram counts, as the baseline model (60.22% accuracy on the new dataset). We also tested several widely-used machine learning methods (logistic regression, Bernoulli Naïve Bayes) trained on the new question representation. The best result of 72.38% accuracy (micro) was achieved with the CNN model. We also ran experiments on pertinent feature selection with a simple Multinomial Naïve Bayes classifier, using word features only, Add-1 smoothing and no strategy for out-of-vocabulary words. Surprisingly, the setting with top-1200 informative word features (by PPMI) and equal priors achieved only slightly lower accuracy, 70.72%, which also beats the baseline by a large margin.
The paper investigates several techniques for hypernymy extraction from a large collection of dictionary definitions in Russian. First, definitions from different dictionaries are clustered, then single words and multiwords are extracted as hypernym candidates. A classification-based approach on pre-trained word embeddings is implemented as a complementary technique. In total, we extracted about 40K unique hypernym candidates for 22K word entries. Evaluation showed that the proposed methods applied to a large collection of dictionary data are a viable option for automatic extraction of hyponym/hypernym pairs. The obtained data is available for research purposes.
Process mining deals with various types of formal models. Some of them are used at intermediate stages of synthesis and analysis, whereas others are the desired goals themselves. Transition systems (TS) are widely used in both scenarios. Process discovery, which is a special case of the synthesis problem, tries to find patterns in event logs. In this paper, we propose a new approach to the discovery problem based on recurrent neural networks (RNN). Here, an event log serves as a training sample for a neural network; the algorithm extracts RNN's internal state as the desired TS that describes the behavior present in the log. Models derived by the approach contain all behaviors from the event log (i.e. are perfectly fit) and vary in simplicity and precision, the key model quality metrics. One of the main advantages of the neural method is the natural ability to detect and merge common behavioral parts that are scattered across the log. The paper studies the proposed method, its properties and possible cases where the application of this approach is sensible as compared to other methods of TS synthesis.