Claudia-Lavinia Ignat

Quality Assessment of Wikipedia Articles without Feature Engineering

As Wikipedia became the largest human knowledge repository, quality measurement of its articles received a lot of attention during the last decade. Most research efforts focused on classification of Wikipedia articles quality by using a different feature set. However, so far, no “golden feature set” was proposed. In this paper, we present a novel approach for classifying Wikipedia articles by analysing their content rather than by considering a feature set. Our approach uses recent techniques in natural language processing and deep learning, and achieved a comparable result with the state-of-the-art.

A CRDT Supporting Selective Undo for Collaborative Text Editing

Undo is an important feature of editors. However, even after over two decades of active research and development, support of undo for real-time collaborative editing is still very limited. We examine issues concerning undo in collaborative text editing and present an approach using a layered commutative replicated data type CRDT. Our performance study shows that it provides sufficient responsiveness to the end users.

Performance of real-time collaborative editors at large scale: User perspective

Real-time collaborative editing allows multiple users to edit a shared document at the same time. It received a lot of attention from both industry and academia and gained in popularity due to the wide availability of free services such as Google Docs. While these collaborative editing systems were initially used in scenarios involving only a small set of users such as for writing a research article, nowadays we notice a change in the scale from several users to communities of users. Group note taking during lectures or conferences is an emerging practice. An important measure of performance of real-time collaborative editing systems is delay. Delays exist between the execution of one user modification and the visibility of this modification to the other users. They can be caused by network physical communication media, complexity of consistency maintenance algorithms and system architecture. Some user studies have shown that delay affects group performance in collaborative editing. In this paper, we measure delays in popular real-time collaborative editing systems such as Google Docs and Etherpad and we study whether these systems could cope with large scale settings from a user perspective. Our results show that these systems are not yet ready for large-scale collaborative activities as either they reject new users connection or a high delay appears when facing an increasing number of users or their typing speeds in the same shared document.

How Do User Groups Cope with Delay in Real-Time Collaborative Note Taking

A property of general interest of real-time collaborative editors is delay. Delays exist between the execution of one user’s modification and the visibility of this modification to the other users. Such delays are in part fundamental to the network, as well as arising from the consistency maintenance algorithms and underlying architecture of collaborative editors. Existing quantitative research on collaborative document editing does not examine either concern for delay or the efficacy of compensatory strategies. We studied an artificial note taking task in French where we introduced simulated delay. We found out a general effect of delay on performance related to the ability to manage redundancy and errors across the document. We interpret this finding as a compromised ability to maintain awareness of team member activity, and a reversion to independent work. Measures of common ground in accompanying chat indicate that groups with less experienced team members attempt to compensate for the effect of delay. In contrast, more experienced groups do not adjust their communication in response to delay, and their performance remains sensitive to the delay manipulation.

An end-to-end learning solution for assessing the quality of Wikipedia articles

Wikipedia is considered as the largest knowledge repository in the history of humanity and plays a crucial role in modern daily life. Assigning the correct quality class to Wikipedia articles is an important task in order to provide guidance for both authors and readers of Wikipedia. The manual review cannot cope with the editing speed of Wikipedia. An automatic classification is required to classify the quality of Wikipedia articles. Most existing approaches rely on traditional machine learning with manual feature engineering, which requires a lot of expertise and effort. Furthermore, it is known that there is no general perfect feature set because information leak always occurs in feature extraction phase. Also, for each language of Wikipedia, a new feature set is required. In this paper, we present an approach relying on deep learning for quality classification of Wikipedia articles. Our solution relies on Recurrent Neural Networks (RNN) which is an end-to-end learning technique that eliminates disadvantages of feature engineering. Our approach learns directly from raw data without human intervention and is language-neutral. Experimental results on English, French and Russian Wikipedia datasets show that our approach outperforms state-of-the-art solutions.

Handling Disturbance and Awareness of Concurrent Updates in a Collaborative Editor

When people work collaboratively on a shared document, they have two contradictory requirements on their editors that may affect the efficiency of their work. On the one hand, they would like to know what other people are currently doing on a particular part of the document. On the other hand, they would like to focus their attention on their own current work, with as little disturbance from the concurrent activities as possible. We present some features that help the user handle disturbance and awareness of concurrent updates. While collaboratively editing a shared document with other people, a user can create a focus region. The user can concentrate on the work in the region without being interfered with the concurrent updates of the other people. Occasionally, the user can preview the concurrent updates and select a number of these updates to be integrated into the local copy. We have implemented a collaborative editing subsystem in the GNU Emacs (https://www.gnu.org/software/emacs) text editor with the described features.

Quality assessment of wikipedia articles: a deep learning approach by Quang Vinh Dang and Claudia-Lavinia Ignat with Martin Vesely as coordinator

Wikipedia is indeed a very important knowledge sharing platform. However, since its start in 2001, the quality of Wikipedia is questioned because its content is created potentially by everyone who ...Wikipedia is indeed a very important knowledge sharing platform. However, since its start in 2001, the quality of Wikipedia is questioned because its content is created potentially by everyone who can access the Internet. Currently, the quality of Wikipedia articles is assessed by human judgement. The method is not scalable up to huge size and fast changing speed of Wikipedia today. An automatic quality classifier for Wikipedia articles is required to support user to choose high quality articles for reading and to notify authors for improving their products. While other existing approaches are based on manually predefined specific feature set, we present our approach of using deep learning to automatically represent Wikipedia articles for quality classification.

An Analysis of Merge Conflicts and Resolutions in Git-Based Open Source Projects

Version control systems such as Git support parallel collaborative work and became very widespread in the open-source community. Whilst Git offers some very interesting features, resolving conflicts that arise during synchronisation of parallel changes is a time-consuming task. In this paper we present an analysis of concurrency and conflicts in official Git repository of four projects: Rails, IkiWiki, Samba and Linux Kernel. We analyse the collaboration process of these projects at specific periods revealing how change integration and conflict rates vary during project development life-cycle. We also analyse how often users decide to rollback to previous document version when the integration process generates conflicts. Finally, we discuss the mechanism adopted by Git to consider changes made on two continuous lines as conflicting.

Trusternity: Auditing Transparent Log Server with Blockchain

Public key server is a simple yet effective way of key management in secure end-to-end communication. To ensure the trustworthiness of a public key server, transparent log systems such as CONIKS employ a tamper-evident data structure on the server and a gossiping protocol among clients in order to detect compromised servers. However, due to lack of incentive and vulnerability to malicious clients, a gossiping protocol is hard to implement in practice. Meanwhile, alternative solutions such as EthIKS are not scalable. This paper presents Trusternity, an auditing scheme relying on Ethereum blockchain that is easy to implement, scalable and inexpensive to operate.

Blockchain-Based Auditing of Transparent Log Servers

Public key server is a simple yet effective way of key management in secure end-to-end communication. To ensure the trustworthiness of a public key server, CONIKS employs a tamper-evident data structure on the server and a gossiping protocol among clients in order to detect compromised servers. However, due to lack of incentive and vulnerability to malicious clients, a gossiping protocol is hard to implement in practice. Meanwhile, alternative solutions such as EthIKS are too costly. This paper presents Trusternity, an auditing scheme relying on Ethereum blockchain that is easy to implement, inexpensive to operate and resilient to malicious clients. We also conduct an empirical study of system behaviour in face of attacks and propose a lightweight anomaly detection algorithm to protect clients against such attacks.