Mikhail Alexandrov

Generating overview timelines for major events in an RSS corpus

Really simple syndication (RSS) is becoming a ubiquitous technology for notifying users of new content in frequently updated web sites, such as blogs and news portals. This paper describes a feature-based, local clustering approach for generating overview timelines for major events, such as the tsunami tragedy, from a general-purpose corpus of RSS feeds. In order to identify significant events, we automatically (1) selected a set of significant terms for each day; (2) built a set of (term–co-term) pairs and (3) clustered the pairs in an attempt to group contextually related terms. The clusters were assessed by 10 people, finding that the average percentage apparently representing significant events was 68.6%. Using these clusters, we generated overview timelines for three major events: the tsunami tragedy, the US election and bird flu. The results indicate that our approach is effective in identifying predominantly genuine events, but can only produce partial timelines.

A Survey of Multilingual Event Extraction from Text

The ability to process multilingual texts is important for the event extraction systems, because it not only completes the picture of an event, but also improves the algorithm performance quality. The present paper is a partial overview of the systems that cover this functionality. We focus on language-specific event type identification methods. Obtaining and organizing this knowledge is important for our further experiments on mono- and multilingual detection of socio-political events.

Constructing empirical models for automatic dialog parameterization

Automatic classification of dialogues between clients and a service center needs a preliminary dialogue parameterization. Such a parameterization is usually faced with essential difficulties when we deal with politeness, competence, satisfaction, and other similar characteristics of clients. In the paper, we show how to avoid these difficulties using empirical formulae based on lexical-grammatical properties of a text. Such formulae are trained on given set of examples, which are evaluated manually by an expert(s) and the best formula is selected by the Ivakhnenko Method of Model Self-Organization. We test the suggested methodology on the real set of dialogues from Barcelona railway directory inquiries for estimation of passengers politeness.

Forming a Representative Sample in the Analysis of Large Data

Russian Abstract: Препринт содержит результаты исследований сотрудников Международной лаборатории математических методов исследования социальных сетей в 2016 г по одному из направлений работы лаборатории, связанному с обработкой больших данных. English Abstract: The working paper contains the results of research of the employees of the International Laboratory of Mathematical Methods for the Study of Social Networks in 2016, according to one of the directions of the laboratorys work connected with the processing of big data.

Creating Collections of Descriptors of Events and Processes Based on Internet Queries

Search queries to Internet are a real reflection of events and processes that happen in the informative society. Moreover, the recent research shows that search queries can be an effective tool for the analysis and forecast of these events and processes. In the paper, we present our experience in creating databases of descriptors (queries and their combinations) to be used in real problems. An example related to the analysis and forecast of regional economy illustrates an application of the mentioned descriptors. The paper is intended for those who use or plan to use Internet queries in their applied research and practical applications.

Selection of Statistically Representative Subset from a Large Data Set

Selecting a representative subset of objects is one of the effective ways for processing large data sets. It concerns both automatic time-consuming algorithms and manual study of object properties by experts. ‘Representativity’ is considered here in a narrow sense as the equality of the statistical distributions of objects parameters for the subset and for the whole set. We propose a simple method for the selection of such a subset based on testing complex statistical hypotheses including an artificial hypothesis to avoid ambiguity. We demonstrate its functionality on two data sets, where one is related to the companies of mobile communication in Russia and the other – to the intercity autobuses communication in Peru.

Multilingual Protest Event Data Collection with GATE

Protest event databases are key sources that sociologists need to study the collective action dynamics and properties. This paper describes a finite-state approach to protest event features collection from short texts (news lead sentences) in several European languages (Bulgarian, French, Polish, Russian, Spanish, Swedish) using the General Architecture for Text Engineering (GATE). The results of the annotation performance evaluation are presented.

A Modified Tripartite Model for Document Representation in Internet Sociology

Seven years ago Peter Mika (Yahoo! Research) proposed a tripartite model of actors, concepts and instances for document representation in the study of social networks. We propose a modified model, where instead of document authors we consider textual mentions of persons and institutions as actors. This representation proves to be more appropriate for the solution of a range of Internet Sociology tasks. In the paper we describe experiments with the modified model and provide some background on the tools that can be used to build it. The model is tested on the experimental corpora of Russian news (educational domain). The research reflects the pilot study findings.

Forecasting Euro/Dollar Rate with Forex News

In the paper we build classifiers of texts reflecting opinions of currency market analysts about euro/dollar rate. The classifiers use various combinations of classes: growth, fall, constancy, not-growth, not-fall. The process includes term selection based on criterion of word specificity and model selection using technique of inductive modeling. We shortly describe our tools for these procedures. In the experiments we evaluate quality of classifiers and their sensibility to term list. The results proved to be positive and therefore the proposed approach can be a useful addition to the existing quantitative methods. The work has a practical orientation.

Performance of Inductive Method of Model Self-Organization with Incomplete Model and Noisy Data

Inductive method of model self-organization (IMMSO) developed in 80s by A. Ivakhnenko is an evolutionary machine learning algorithm, which allows selecting a model of optimal complexity that describes or explains a limited number of observation data when any a priori information is absent or is highly insufficient. In this paper, we study the performance of IMMSO to reveal a model in a given class with different volumes of data, contributions of unaccounted components, and levels of noise. As a simple case study, we consider artificial observation data: the sum of a quadratic parabola and cosine; model class under consideration is a polynomial series. The results are interpreted in the terms of signal-noise ratio.