IMCAFS

 Home

threat detection of twitter using deep neural network

Posted by lipsius at 2020-02-24
all

Core Express

Using deep neural network to detect the threat of twitter;

Affinity: a similarity comparison system for potential users of SMS data;

The emergence of network bifurcations caused by entanglement;

Market structure of online dating in American cities;

The semi supervision chart shows the growing party relations in the US Congress;

In the repeated prisoner's dilemma game, when there is observation error, the strategy of forcing linear return relationship is adopted;

The symbiosis of arXiv with physical preprints and journal reviews: a model;

Network social algorithm analysis of collective choice on behalf of the Committee;

Comments on the complementarity of renewable energy: concepts, indicators, applications and future research directions;

New transition and Bellerophon state of coupling Stuart Landau oscillator;

Crying wolf effect in evacuation: a game theory method;

Unsupervised twitter user position detection;

Adjustable centrality based on eigenvector in multi-channel network and time network;

Integrated deep learning model for drug abuse detection in sparse twitter sphere;

A comparative study of hierarchical navigable small world map;

According to the national consumption data, the country specific space heating threshold temperature is estimated;

Using depth neural network

Threat detection on twitter

Original title:

Cyberthreat Detection from Twitter using Deep Neural Networks

Address:

http://arxiv.org/abs/1904.01127

Author:

Nuno Dionísio, Fernando Alves, Pedro M. Ferreira, Alysson Bessani

Absrtact: in order to prepare for network attack, most organizations use security information and event management system to monitor their infrastructure. These systems depend on the latest updates, patches and threats provided by cyber threat intelligence sources. Open source intelligent platform, namely social media networks such as twitter, can gather a large number of resources related to network security. In order to deal with this kind of information flow, we need scalable and efficient tools, which can identify and summarize the relevant information of specified assets.

This paper introduces the processing flow of a new tool, which uses deep neural network to process the network security information received from twitter. Convolutional neural networks identify tweets that contain security related information about assets in the IT infrastructure. Then, bi-directional short and long term memory networks extract named entities from these tweets to form security alerts or fill in compromise indicators. In three case study infrastructures, the proposed pipeline achieved an average of 94% true positive and 91% true negative for classification tasks and an average of 92% F1 for named entity recognition tasks.

Affinity: a number for SMS

Comparison system of similarity of potential users based on

Original title:

affinity: A System for Latent User Similarity Comparison on Texting Data

Address:

http://arxiv.org/abs/1904.01897

Author:

Tobias Eichinger, Felix Beierle, Sumsam Ullah Khan, Robin Middelanis, Veeraraghavan Sekar, Sam Tabibzadeh

Absrtact: in the field of social network services, it is common to find similar users based on profile data. Smartphones contain sensor and personal context data that can be used for user analysis. However, a huge source of personal data, short message data, has hardly been used for user analysis. We think there are three reasons: first, private SMS data cannot be shared because of its intimate characteristics. Secondly, the definition of privacy appropriate similarity measure is nontrivial. Thirdly, the quality of similarity measurement of text message data representing potential infinite topic set is nontrivial.

In order to overcome these obstacles, we propose affinity, which is a reliable and effective system to evaluate the similarity between users' text message history in the way of privacy protection. The private SMS data is retained on the user's device, and the data used for comparison is compared in a potential format that neither allows the reconstruction of the comparison word nor any original private plain text. We evaluated our approach by calculating the similarities between the twitter histories of 60 U.S. senators. The resulting similarity network achieves an average accuracy of 85.0% in Party classification tasks.

The emergence of network bifurcations caused by entanglement

Original title:

Emergence of Network Bifurcation Triggered by Entanglement

Address:

http://arxiv.org/abs/1609.03299

Author:

Xi Yong, Man-Hong Yung, Xue-Ke Song, Xun Gao, Angsheng Li

Abstract: in many nonlinear systems, such as plasma oscillation, boson condensation, chemical reaction, even predator-prey oscillation, the coarse-grained dynamics is controlled by the equation with antisymmetric transition, which is called antisymmetric Lotka Volterra (ALV) equation. In this work, we prove the existence of a new bifurcation mechanism for ALV equation, in which the equilibrium state can be greatly changed by turning the stability of a pair of fixed points. As an application, we pay attention to the influence of bifurcation mechanism on the evolutionary network; we find that the bifurcation point can be determined quantitatively by micro quantum entanglement.

The equilibrium state can be transformed from one type of global population cohesion to another supporting the global cooperation of homogeneous networks. In other words, our results show that there is a kind of multibody system, in which the macroscopic properties and a certain amount of microscopic entanglements are invariable, but once the entanglements exceed the critical value, they will suddenly change. In addition, the numerical evidence provided by us shows that the occurrence of bifurcation is robust to the change of network topology, and the critical value is very consistent with our theoretical prediction. These results show that the bifurcation mechanism is ubiquitous in many physical systems except for the evolutionary networks.

Market structure of online dating in American cities

Original title:

Structure of online dating markets in US cities

Address:

http://arxiv.org/abs/1904.01050

Author:

Elizabeth E. Bruch, M. E. J. Newman

Absrtact: by analyzing the interaction of millions of users in large online dating websites, we apply the recently developed network analysis method to the message mode of users' exchange, and study the structure of heterosexual dating market in the United States. Our analysis shows that the strongest driving force of romantic interaction at the national level is simple geographical proximity, but at the local level, other demographic factors also play a role.

We found that the dating market in each city is divided into sub markets according to age and race. There is a big difference in the gender ratio between different submarkets. Young submarkets have more males than the elderly, and fewer females. Minorities, especially women, are also younger than the average in older submarkets, and our analysis reveals how this racial stratification is determined by the information transmission of men and women. Our research shows how Internet technology can be applied to online interaction to reveal the overall impact of individual behavior on social structure.

Semi supervised icon showing the United States

Growing partisanship in Congress

Original title:

Semi-supervised graph labelling reveals increasing partisanship in the United States Congress

Address:

http://arxiv.org/abs/1904.01153

Author:

Max Glonek, Jonathan Tuke, Lewis Mitchell, Nigel Bean

Absrtact: Iconography is a key activity of network science, which has a wide range of practical applications and is closely related to other network science tasks, such as community detection and clustering. Although there are a lot of work in unsupervised and supervised tagging algorithms, the random walk based algorithms need to be further explored, especially considering their relevance to social and political networks.

This work improves and expands a new semi supervised icon recording method, glass method, which can accurately calculate the absorption probability of random walks on connected graphs. In this method, the graph is accurately described as a discrete-time Markov chain, and the labeled nodes are regarded as absorbing states. This method is applicable to the data of the roll call table of 42 meetings of the house of Representatives and the Senate of the United States from 1935 to 2019. The analysis of 84 result political networks shows that glass has strong and consistent performance when estimating the unlabeled nodes in the graph, and reveals the important trend of increasing party relations in the United States Congress.

Observation in repeated prisoner's dilemma game

Strategy of forcing linear return relationship in case of error

Original title:

Strategies that enforce linear payoff relationships under observation errors in Repeated Prisoner’s Dilemma game

Address:

http://arxiv.org/abs/1904.01208

Author:

Azumi Mamiya, Genki Ichinose

Absrtact: repeated game theory analyzes the long-term relationship of interactive players, and mathematically reveals the conditions of how to achieve cooperation, which is impossible to achieve in one-off game. In the RPD game without mistakes, the ZD strategy allows players to unilaterally set the linear relationship between the player's own earnings and the opponent's earnings, regardless of the strategy implemented by the opponent. In contrast, unconditional strategies such as alld and ALLC also unilaterally set up linear payment relationships.

Mistakes often happen between players in the real world. However, little is known about the existence of such strategies with errors in RPD games. Here, we analytically search for strategies to enforce linear payment relationships under observation errors in RPD games. As a result, we find that even in the case of observation errors, the only set of policies to enforce the linear payment relationship is ZD strategy or unconditional strategy, which are confirmed numerically. In addition, we derive the feasible range of the expected return of the unconditional strategy.

ArXiv and physical preprint and date

Symbiosis of journal review articles: a model

Original title:

arXiv and the Symbiosis of Physics Preprints and Journal Review Articles: A Model

Address:

http://arxiv.org/abs/1904.01470

Author:

Brian Simboli

Abstract: This paper recommends a publishing model which is helpful to achieve the goal of Physics Publishing Reform. It distinguishes two complementary needs in academic exchanges. Preprints are becoming more and more important in science, and they are appropriate tools for claiming priority in discovery and eliciting feedback that helps version control. However, traditional periodical publishing should focus on providing synthesis in the form of covering periodicals, which play the same role as commentary articles.

Collective selection on behalf of the Committee

Analysis of network social algorithm based on

Original title:

Analysis of a networked social algorithm for collective selection of a committee of representatives

Address:

http://arxiv.org/abs/1904.01602

Author:

Alexis R. Hernandez, Carlos Gracia-Lazaro, Edgardo Brigatti, Yamir Moreno

Abstract: Hern andez et al. A voting rule supported by trust based social networks is introduced, in which the instructions of possible representatives are based on personal opinions. Personal donation is not only a simple counting of votes, but also based on the subject voting. These mechanisms create a high level of representation of the selected committees and reduce the possibility of sponsorship relationships. By combining the integrity and perception of individuals, we discuss the credibility of the resulting Committee here. Our results show that the voting rules provide a high degree of representation for small committees with a high degree of integrity. In addition, the voting system shows the robustness to the strategy of voting algorithm and the unreal application.

Review on the complementarity of renewable energy

Ideas, indicators, applications and future research directions

Original title:

A review on the complementarity of renewable energy sources: concept, metrics, application and future research directions

Address:

http://arxiv.org/abs/1904.01667

Author:

J. Jurasz, F.A. Canales, A. Kies, M. Guezgouz, A. Beluco

Abstract: it is expected and observed in the region that the extensive deployment of renewable energy will soon cover energy demand. However, weather and climate driven energy is characterized by significant spatial and temporal changes. One of the common solutions to overcome the mismatch between the supply and demand of renewable energy is to mix two or more energy sources in a single power station (such as wind solar, solar water or solar wind hydro). The operation of hybrid energy is based on the complementary nature of renewable energy.

Considering the growing importance of such systems and more and more research activities in this field, this paper reviews the research on the complementary effects of space-time, space and time between research, analysis, quantification and utilization of renewable energy. Firstly, the review briefly summarizes the existing research papers, makes a detailed definition of the main concepts, summarizes the current research direction, and ends the future research activities. The evaluation provides temporal and spatial information on the study of complementary concepts.

Coupled Stuart Landau oscillator

New transitions and Bellerophon states of

Original title:

Novel transition and Bellerophon state in coupled Stuart-Landau oscillators

Address:

http://arxiv.org/abs/1904.01776

Author:

Jiameng Zhang, Xue Li, Yong Zou, Shuguang Guan

Abstract: we study the synchronization in Stuart Landau oscillator system with frequency weighted coupling. For three typical single peak frequency distributions, Lorentz, trigonometric and uniform, we find that the first-order transition occurs when the frequency distribution is relatively compact, while the synchronous transition is continuous when the frequency distribution is relatively wide. In these two cases, there is a bellerophene state between the incoherent state and the synchronous state.

It is worth noting that we reveal the new transition behavior of the coupled oscillator with amplitude, that is, the state of the Bellerophon state actually consists of two stages. In the first stage, the oscillator achieves chaotic phase synchronization, while in the second stage, the oscillator forms periodic phase synchronization. Our results show that the Bellerophon state also exists in a coupled oscillator with amplitude dynamics.

Crying wolf effect in evacuation

A game theory method

Original title:

The Cry Wolf Effect in Evacuation: a Game-Theoretic Approach

Address:

http://arxiv.org/abs/1904.01963

Author:

Alexandros Rigos, Enrico Ronchi, Erik Mohlin

Abstract: in today's terrorism and security centered world, evacuation emergencies, drills and false alarms are becoming more and more common. In an emergency, the authorities' compliance with the evacuation order can play a key role in the outcome of the emergency. If the evacuees experience repeated emergencies, it may be a false alarm (e.g. evacuation exercise, false bomb threat, etc.) or an actual threat. The crying wolf effect of Aesop (repeated false alarm reduces order compliance) will seriously affect the possibility of his / her evacuation. In order to analyze this key unsolved evacuation research problem, a game theory method is proposed.

Game theory is used to explore the best reaction between evacuees and authorities. In the proposed model, authorities get a signal of whether there is a threat and decide whether to order an evacuation. After receiving the evacuation order, the evacuees decide whether to stay or leave according to their later beliefs, which are updated according to the actions of the authorities. The optimal response is derived, and sequential equilibrium and perfect Bayesian equilibrium are used as solution concepts (refining equilibrium with intuitive criteria). The model results highlight the benefits of the announced evacuation exercise and show that increasing the accuracy of threat detection can prevent a large number of inefficiencies associated with wolf wolf effects.

Unsupervised twitter user position detection

Original title:

Unsupervised User Stance Detection on Twitter

Address:

http://arxiv.org/abs/1904.02000

Author:

Kareem Darwish, Peter Stefanov, Michaël J. Aupetit, Preslav Nakov

Absrtact: we propose a very effective unsupervised method to detect Twitter users' positions on controversial topics. In particular, we use dimensionality reduction to project users into low dimensional space, and then cluster them, which enables us to find core users representing different positions. Compared with the most advanced methods, our method has three main advantages. These methods are based on supervised or semi supervised classification.

First, we don't need to tag users in advance: instead, we create clusters, which are easier to tag manually later, for example, in seconds or minutes rather than hours. Second, domain or topic level knowledge is not required to specify the relevant positions (labels) or to make actual labels. Third, our method is robust in the face of data skew, for example, when some users or some sites have a larger representation in the data.

We try different combinations of user similarity feature, data set size, dimension reduction method and clustering algorithm to determine the most effective and efficient combination across three different data sets (English and Turkish). Our best combination of effectiveness and efficiency uses the forwarded account as a function, umap for dimensionality reduction, and mean shift for clustering, and generates a small number of high-quality user groups, usually only 2-3, more than 98% purity. In addition, our method is robust to the variation of parameter values and random initialization.

In multiplex network and time network

Adjustable centrality based on Eigenvector

Original title:

Tunable Eigenvector-Based Centralities for Multiplex and Temporal Networks

Address:

http://arxiv.org/abs/1904.02059

Author:

Dane Taylor, Mason A. Porter, Peter J. Mucha

Absrtact: the importance of representing nodes in social, biological, information and technology networks is the core theme of network science and data science. We propose a linear algebra framework, which generalizes the center based on eigenvectors - including PageRank - to provide a general framework for two popular multi-layer network classes: the multi-channel network with layers encoding different types of relations and the change of time network of relations among them.

Our method involves the study of union, marginal and conditional hypercentricity, which can be obtained from the main eigenvectors of hypercentric matrices [Taylor et al. , 2017], which couples the central matrix associated with each network layer. We extend this previous work (limited to time networks with layers passing through undirected, time adjacent coupling) by allowing layers to be coupled via (possibly asymmetric) inter layer adjacency matrix, where each entry gives the coupling between layers T and t '. Our framework provides a unified basis for the centrality analysis of multiple and / or time networks, and reveals the complex dependence of hypercentricity on inter layer coupling topology.

We extend tilde BF a by coupling strength Omega ge0, and develop singular perturbation theory to infty for the limit of weak (omega to 0 ^ +) and strong coupling (omega), revealing the interesting dependence on the main eigenvectors of tilde BF a. We apply the framework to two empirical datasets: a multi network representation of air transport in Europe, and a time network to encode the graduation and employment of mathematicians in American universities and colleges of mathematics.

Drugs in sparse twitter sphere

An integrated deep learning model for abuse detection

Original title:

An Ensemble Deep Learning Model for Drug Abuse Detection in Sparse Twitter-Sphere

Address:

http://arxiv.org/abs/1904.02062

Author:

Han Hu, NhatHai Phan, James Geller, Stephen Iezzi, Huy Vo, Dejing Dou, Soon Ae Chun

Abstract: with the aggravation of drug abuse in the United States, many researches mainly use social media data (such as posts on twitter) to study drug abuse related activities, using machine learning as a powerful tool for text classification and filtering. However, given the broad themes of Twitter users, tweets related to drug abuse in most data sets are rare. This kind of unbalanced data is still the main problem of constructing an effective tweeter classifier, especially for the study of slang terms related to abuse.

In this study, we design a set depth learning model to solve this problem. This model uses word level and character level functions to classify tweets related to abuse. Reporting experiments on twitter datasets, where we can configure the percentage of two categories (abuse and non abuse) to simulate data imbalances of different magnitude. The results show that our integrated deep learning model performs better than the set of traditional machine learning models, especially on severely unbalanced data sets.

A comparative study of hierarchical navigable small world map

Original title:

A Comparative Study on Hierarchical Navigable Small World Graphs

Address:

http://arxiv.org/abs/1904.02077

Author:

Peng-Cheng Lin, Wan-Lei Zhao

Absrtact: since the release of source code two years ago, hierarchical navigable small world (hnsw) map has become more and more popular in large-scale nearest neighbor search tasks. The attraction of this method lies in its superior performance over most nearest neighbor search methods and its versatility for various distance measurements.

In this paper, several comparative studies are made on this search method. The role of hierarchy in hnsw and the role of hnsw graph itself are studied. We find that the hierarchical structure in hnsw can not achieve the "better log complexity scale", especially in high-dimensional data. In addition, we find that after graph diversification, with the support of planar k-NN graph, we can achieve high search speed and efficiency similar to hnsw. Finally, we point out that the difficulties faced by most graph based search methods are directly related to the curse of dimension. Like other chart based methods, hnsw cannot solve this problem.

Estimating countries based on national consumption data

Home specific space heating threshold temperature

Original title:

Estimating country-specific space heating threshold temperatures from national consumption data

Address:

http://arxiv.org/abs/1904.02080

Author:

Smail Kozarcanin, Gorm Bruun Andresen, Iain Staffell

Abstract: space heating in buildings is becoming a key element in the study of sector coupled energy systems. Data availability limits modeling efforts in the construction industry because most countries do not directly measure heat consumption. Space heating is usually related to the degree of heating days representing the weather using a specific heating threshold temperature, but the approach is different between studies. This study uses extensive and public consumption and weather data to estimate country specific heating threshold temperatures. This allows the capture of national climate and culture specific human behavior in energy system modeling.

National electricity and natural gas consumption data are related to degree day by linear model, and Akaike's information standard is used to define summer in each country when space heating is not required. We found that the heating threshold temperature calculated using daily, weekly and monthly aggregate consumption data was statistically insignificant. Generally speaking, the critical temperature of gas heating is about 15.0 + / - 1.7 ℃ (daily average temperature), while the average temperature of electric heating is 13.4 + / - 2.4 ℃. We found that there was no sign of space heating during June, July and August, even if there was a heating day.

Source: internet science research express

Editor: Meng Jie

Disclaimer: the copyright of arXiv article summary belongs to the original author of the paper, which shall be translated and arranged by myself. Please do not reprint without permission. This series is updated in WeChat official account, "network science research express" (micro signal netsci) and personal blog https://www.complexly.me (RSS subscription).

Express delivery of recent internet scientific papers

Unlock the language patterns in dog behavior

Negative attitude to science comes from overconfidence and other 17 articles

14 accurate "no free lunch" theorems of community testing

10 graph structure data sets for Wikipedia research

24 new heterogeneous graph attention networks

Join Jizhi, complicated together!

QQ group of Jizhi club 877391004

Business cooperation and contribution Reprint: [email protected]

Search official account: Ji Chi Club

Join the Institute without walls

Let the apple hit harder!

© 2023