Search

Applications are one of the most used attack surfaces, and they must be secured at source code level, early in the development phase. Developers' inherited culture of programming preserves the patterns of code writing within big organizations or developers' communities, opening an opportunity to use SAST (Static Application Security Testing) complementary solutions to identify insecure code early in the development phase. We propose an Intermediate Representation, strict enough to maintain the security vulnerabilities patterns as defined by MITRE with the Common Weaknesses Enumeration, at the same time agile enough to not strongly depend on the lexical and syntax structure of the programming language, but following programmers' behavior of writing code. The current research phase uses semantical clustering of instructions (keywords) found in C/C++ programs, based on Word Embeddings, which are transported via the resulting (numerical) Intermediate Representation to the various classifiers for security vulnerability patterns detection. We show that there is a good preservation of security patterns despite the generalization of keywords via semantical clustering. This opens an opportunity for innovation in security vulnerability patterns identification, which is more dependent on the programmers' code writing behavior than the programming language specific structure.

We propose a novel solution for assigning labels to topic models by using multiple weak labelers. The method leverages generative transformers to learn accurate representations of the most important topic terms and candidate labels. This is achieved by fine-tuning pre-trained BART models on a large number of potential labels generated by state of the art non-neural models for topic labeling, enriched with different techniques. The proposed BART-TL model is able to generate valuable and novel labels in a weakly-supervised manner and can be improved by adding other weak labelers or distant supervision on similar tasks.

In this paper, we propose a new corpus and approach to evaluate and detect the irony in online news published in Romanian. Irony detection is based on subjective perceptions, thus the human experience plays an important role in this field. We propose a supervised machine learning system based on a Romanian ironic and non-ironic news corpus, manually annotated, and a Romanian dictionary of commonly used words. We present a comparison of the best results by applying several classification algorithms, such as Naive Bayes, Logistic Regression, Linear SVC, Decision Trees, and Random Forest. We present aspects about irony detection and some language resources, considering a Romanian corpus based on 25,841 non-ironic news items and 14,064 ironic news items collected from online sources, which is used by the proposed irony detection system. Furthermore, we present the results of a sentiment analysis system based on a supervised machine learning approach applied to the 8,128 news items representing our evaluation corpus. The scope of the paper is to present the proposed system and a practical solution for automatic irony detection in Romanian real-world applications, which can provide useful information about online media perception. The best result of the compared systems, after having processed the corpus and applied the filtering method is over 91 %, achieved by the Naive Bayes algorithm.

Sentiment analysis is a process widely used in opinion mining campaigns conducted today. This phenomenon presents applications in a variety of fields, especially in collecting information related to the attitude or satisfaction of users concerning a particular subject. However, the task of managing such a process becomes noticeably more difficult when it is applied in cultures that tend to combine two languages in order to express ideas and thoughts. By interleaving words from two languages, the user can express with ease, but at the cost of making the text far less intelligible for those who are not familiar with this technique, but also for standard opinion mining algorithms. In this paper, we describe the systems developed by our team for SemEval-2020 Task 9 that aims to cover two well-known code-mixed languages: Hindi-English and Spanish-English. We intend to solve this issue by introducing a solution that takes advantage of several neural network approaches, as well as pre-trained word embeddings. Our approach (multlingual BERT) achieves promising performance on the Hindi-English task, with an average F1-score of 0.6850, registered on the competition leaderboard, ranking our team 16th out of 62 participants. For the Spanish-English task, we obtained an average F1-score of 0.7064 ranking our team 17th out of 29 participants by using another multilingual Transformer-based model, XLM-RoBERTa., Comment: Accepted at SemEval-2020, 9 pages, 4 tables

A natural language interface to databases (NLIDB) enables users without technical expertise to easily access information from relational databases. Interest in NLIDBs has resurged in the past years due to the availability of large datasets and improvements to neural sequence-to-sequence models. In this survey we focus on the key design decisions behind current state of the art neural approaches, which we group into encoder and decoder improvements. We highlight the three most important directions, namely linking question tokens to database schema elements (schema linking), better architectures for encoding the textual query taking into account the schema (schema encoding), and improved generation of structured queries using autoregressive neural models (grammar-based decoders). To foster future research, we also present an overview of the most important NLIDB datasets, together with a comparison of the top performing neural models and a short insight into recent non deep learning solutions.

Goal-Oriented Chatbots in fields such as customer support, providing certain information or general help with bookings or reservations, suffer from low performance partly due to the difficulty of obtaining large domain-specific annotated datasets. Given that the problem is closely related to the domain of the conversational agent and the data belonging to a specific domain is difficult to annotate, there have been some attempts at surpassing these challenges such as unsupervised pre-training or transfer learning between different domains. A more thorough analysis of the transfer learning mechanism is justified by the significant improvement of the results demonstrated in the results section. We describe extensive experiments using transfer learning and warm-starting techniques with improvements of more than 5% in relative percentage of success rate in the majority of cases, and up to 10x faster convergence as opposed to training the system without them.

In this paper, we propose a new approach to evaluate online feelings, emotions or opinions that allows text classification with some polarity and produce a solution for some relevant sentiment analysis challenges that improves the reliability of sentiment analysis performed. Thus, we propose a semi-supervised machine learning system, based on a taxonomy of emotionally charged words with three classes, taking into account the neutral polarity class, but also a comparison of results by applying several classification algorithms, such as Naive Bayes, Decision Trees, Support Vector Machines and our proposed approach. We present aspects about natural language complexity from the sentiment perspective and some language resources, considering a Romanian corpus based on 25,841 news and a Romanian language dictionary containing 42,497 words, which can be used by sentiment analysis systems. The scope of the paper is to identify the utility and usability of multiple data sets in a real-world application, to get relevant results about online comments perception that can provide valuable future perspectives for a nation’s social and even political issues. The best result of our implemented system, after having processed the corpus, is over 82%, compared to classical machine learning classification methods with a best score of 70%.

This paper presents a system developed for detecting sexual predators in online chat conversations using a two-stage classification and behavioral features. A sexual predator is defined as a person who tries to obtain sexual favors in a predatory manner, usually with underage people. The proposed approach uses several text categorization methods and empirical behavioral features developed especially for the task at hand. After investigating various approaches for solving the sexual predator identification problem, we have found that a two-stage classifier achieves the best results. In the first stage, we employ a Support Vector Machine classifier to distinguish conversations having suspicious content from safe online discussions. This is useful as most chat conversations in real life do not contain a sexual predator, therefore it can be viewed as a filtering phase that enables the actual detection of predators to be done only for suspicious chats that contain a sexual predator with a very high degree. In the second stage, we detect which of the users in a suspicious discussion is an actual predator using a Random Forest classifier. The system was tested on the corpus provided by the PAN 2012 workshop organizers and the results are encouraging because, as far as we know, our solution outperforms all previous approaches developed for solving this task.

In recent years, there have been several advancements in question answering systems. These were achieved both due to the availability of a greater number of datasets, some of them significantly larger in size than any of the existing corpora, and to the recent advancements in deep learning for text classification. In this paper, we explore the improvements achieved by employing neural networks using the features computed by a string kernel for sentence/answer selection. We have validated this approach using two different standard corpora used as benchmarks in question answering and we have found a significant improvement over string kernels and other unsupervised methods for sentence selection.

While semantic parsing has been an important problem in natural language processing for decades, recent years have seen a wide interest in automatic generation of code from text. We propose an alternative problem to code generation: labelling the algorithmic solution for programming challenges. While this may seem an easier task, we highlight that current deep learning techniques are still far from offering a reliable solution. The contributions of the paper are twofold. First, we propose a large multi-modal dataset of text and code pairs consisting of algorithmic challenges and their solutions, called AlgoLabel. Second, we show that vanilla deep learning solutions need to be greatly improved to solve this task and we propose a dual text-code neural model for detecting the algorithmic solution type for a programming challenge. While the proposed text-code model increases the performance of using the text or code alone, the improvement is rather small highlighting that we require better methods to combine text and code features.

Question answering is one of the most important and difficult applications at the border of information retrieval and natural language processing, especially when we talk about complex science questions which require some form of inference to determine the correct answer. In this paper, we present a two-step method that combines information retrieval techniques optimized for question answering with deep learning models for natural language inference in order to tackle the multi-choice question answering in the science domain. For each question-answer pair, we use standard retrieval-based models to find relevant candidate contexts and decompose the main problem into two different sub-problems. First, assign correctness scores for each candidate answer based on the context using retrieval models from Lucene. Second, we use deep learning architectures to compute if a candidate answer can be inferred from some well-chosen context consisting of sentences retrieved from the knowledge base. In the end, all these solvers are combined using a simple neural network to predict the correct answer. This proposed two-step model outperforms the best retrieval-based solver by over 3% in absolute accuracy., 8 pages, 2 figures, 8 tables, accepted at IJCNN 2019

Recent advances in neural code generation have incorporated syntax to improve the generation of the target code based on the user's request in natural language. We adapt the model of [1] to the Natural Language Interface to Databases (NLIDB) problem by taking into account the database schema. We evaluate our model on the recently introduced WIKISQL and SENLIDB datasets. Our results show that the syntax-guided model outperforms a simple sequence-to-sequence (SEQ2SEQ) baseline on WIKISQL, but has trouble with the SENLIDB dataset due to its complexity.

Chat conversations between more than two participants are often used in Computer Supported Collaborative Learning (CSCL) scenarios because they enhance collaborative knowledge sharing and sustain creativity. However, multi-participant chats are more difficult to follow and analyze due to the complex ways in which different discussion threads and topics can interact. This paper introduces a novel method based on neural networks for detecting implicit links that uses features computed with string kernels and word embeddings. In contrast to previous experiments with an accuracy of 33%, we obtained a considerable increase to 44% for the same dataset. Our method represents an alternative to more complex deep neural networks that cannot be properly used due to overfitting on limited training data.

This paper presents a research on using Word2Vec for determining implicit links in multi-participant Computer-Supported Collaborative Learning chat conversations. Word2Vec is a powerful and one of the newest Natural Language Processing semantic models used for computing text cohesion and similarity between documents. This research considers cohesion scores in terms of the strength of the semantic relations established between two utterances, the higher the score, the stronger the similarity between two utterances. An implicit link is established based on cohesion to the most similar previous utterance, within an imposed window. Three similarity formulas were used to compute the cohesion score: an unnormalized score, a normalized score with distance and Mihalcea's formula. Our corpus of conversations incorporated explicit references provided by authors, which were used for validation. A window of 5 utterances and a 1-minute time frame provided the highest detection rate both for exact matching and matching of a block of continuous utterances belonging to the same speaker. Moreover, the unnormalized score correctly identified the largest number of implicit links.

NLCP - The Natural Language Compiler - is a new programming environment, which features a strong type system and support for a natural language interface. Additionally, it provides an interactive interpreter that can assist in the process of designing an algorithm. This is achieved by allowing the user to embed information about the objectives of algorithms within the type system. We provide an overview of the main design considerations behind NLCP and list several simple examples of algorithms translated from natural language statements to code.

This paper details a novel method for grouping together online social networking profiles of the same person extracted from different sources. Name ambiguity arises naturally in any culture due to the popularity of specific names which are shared by a large number of people. This is one of the main problems in people search, which is also multiplied by the number of different data sources that contain information about the same person. Grouping pages from various social networking websites in order to disambiguate between different individuals with the same name is an important task in people search. This allows building a detailed description and a consolidated online identity for each individual. Our results show that given a large enough dataset, neural networks and word embeddings provide the best method to solve this problem.

Progress in statistical paraphrase generation has been hindered for a long time by the lack of large monolingual parallel corpora. In this paper, we adapt the neural machine translation approach to paraphrase generation and perform transfer learning from the closely related task of entailment generation. We evaluate the model on the Microsoft Research Paraphrase (MSRP) corpus and show that the model is able to generate sentences that capture part of the original meaning, but fails to pick up on important words or to show large lexical variation.

In this paper we discuss the detection of concepts for question answering systems using Linked Data. The emphasis will be put on the detection of types and properties. We will present different methods that are currently being used for these tasks and the improvements proposed to improve the performance of QA systems. Both modules have been trained by analyzing a large subset of Wikipedia pages which have a correspondent in the DBpedia [1] knowledge base. The paper also presents the preliminary test results for both modules which show that the proposed methods have a good performance and should be integrated in the next version of our QA system over Linked Data.

This paper analyzes the results obtained by computing the semantic similarity score between pairs of words using different WordNet-based and corpus-based techniques for a custom-built corpus of documents. The authors computed semantic similarity scores for all the pairs of the most frequent words from the documents contained in the corpus, which consisted of multi-party online chat conversations between students engaged in a Computer-Supported Collaborative Learning (CSCL) task. Differences between results provided by the two types of similarity measurement techniques were observed, but also between techniques of the same type. Results were compared to expected results provided by human experts in order to see what similarity measurement technique is better for our custom-built corpus of texts extracted from a CSCL task.

In the context of constantly evolving carry-on technology and its increasing accessibility, namely smart-phones and tablets, a greater need for reliable authentication means comes into sight. The current study offers an alternative solution of uninterrupted testing towards verifying user legitimacy. A continuously collected dataset of 41 users’ touch-screen inputs provides a good starting point into modeling each user’s behavior and later differentiate among users. We introduce a system capable of processing features based on raw data extracted from user-screen interactions and attempting to assign each gesture to its originator. Achieving an accuracy of over 83 %, we prove that this type of authentication system is feasible and that it can be further integrated as a continuous way of disclosing intruders within given mobile applications.

Many historic periods (or events) are remembered by slogans, expressions or words that are strongly linked to them. Educated people are also able to determine whether a particular word or expression is related to a specific period in human history. The present paper aims to establish correlations between significant historic periods (or events) and the texts written in that period. In order to achieve this, we have developed a system that automatically links words (and topics discovered using Latent Dirichlet Allocation) to periods of time in the recent history. For this analysis to be relevant and conclusive, it must be undertaken on a representative set of texts written throughout history. To this end, instead of relying on manually selected texts, the Google Books Ngram corpus has been chosen as a basis for the analysis. Although it provides only word n-gram statistics for the texts written in a given year, the resulting time series can be used to provide insights about the most important periods and events in recent history, by automatically linking them with specific keywords or even LDA topics.

In this paper we present an application for identifying archaisms and neologisms in texts. The application also provides the ability to view graphically the evolution trends of these words for a better interpretation of the results. The presented solution consists of two phases: the learning phase in which we identify the general evolution trends of three categories of words (archaisms, neologisms and common words) and the classification phase in which we label new words with their corresponding category. For both phases, the application requires Internet access because it is using the Google Books N-gram Viewer to generate the images that back up the decisions.

Chat conversations are used for a large range of Computer-Supported Collaborative Learning (CSCL) tasks especially because they allow the creation of multiple conversation threads that run in parallel. Thus, several different topics can be debated at the same time, fostering the exploitation of different ideas and facilitating collaborative knowledge creation. In order to detect these threads, our method proposed to firstly detect the links that arise between the utterances of a conversation. From a computational linguistics perspective, there is a wide variety of different types of links between utterances and there is no mechanism to compute all of them. This paper proposes to explain to what degree semantic similarity measures from Natural Language Processing (NLP) may be used to detect the links that arise between utterances in CSCL chat conversations and which is the effectiveness of applying solely this technique for implicit links identification.

As all natural languages are subject to change over time and as the Web becomes more prevalent, it also constitutes a major source for identifying language evolution. Although these changes affect all linguistic branches ranging from phonetics, lexicon and grammar to semantics and pragmatics, we will focus only on discovering new potential words that entered the Romanian lexicon or alternative forms (lexicalizations) that are frequently used. In this paper we describe the architecture of a system which models the rate of Romanian vocabulary growth based on different statistics gathered by a focused web crawler. In order to validate the proposed system, the paper also presents the main new words identified by the system in online texts written in Romanian.

There are many applications that are incorporating a human appearance and intending to simulate human dialog, but in most of the cases the knowledge of the conversational bot is stored in a database created by a human experts. However, very few researches have investigated the idea of creating a chat-bot with an artificial character and personality starting from web pages or plain text about a certain person. This paper describes an approach to the idea of identifying the most important facts in texts describing the life (including the personality) of an historical figure for building a conversational agent that could be used in middle-school CSCL scenarios.

In this paper we propose a new method for identifying creativity that is based on analyzing a corpus of chat conversations on the same topic and extracting the new ideas expressed by participants. The application is a first step in supporting creativity in online group discussions by highlighting the novel concepts present in conversations new ideas and also by identifying topics that could have become important, if not forgotten during the debates lost ideas.

Vandalism of the content has always been one of the greatest problems for Wikipedia, yet only few completely automatic solutions for solving it have been developed so far. Volunteers still spend large amounts of time correcting vandalized page edits, instead of using this time to improve the quality of the content of articles. The purpose of this paper is to introduce a new vandalism detection system, that only uses natural language processing and machine learning techniques. The system has been evaluated on a corpus of real vandalized data in order to test its performance and justify the design choices. The same expert annotated wikitext, extracted from the encyclopedia's database, is used to evaluate different vandalism detection algorithms. The paper presents a critical analysis of the obtained results, comparing them to existing solutions, and suggests different statistical classification methods that bring several improvements to the task at hand.

The paper presents a system for helping the analysis of Computer-Supported Collaborative Learning chat (instant messenger) sessions, starting from a polyphonic model of the discourse inspired by the dialogistics of Bakhtin. Some theoretical basics of the model are presented, followed by implementation details and validation results.

The wider acceptance and usage of instant messaging (chat) represents one of the consequences of undertaking Computer-Supported Collaborative Learning (CSCL) practices in formal education settings. However, the difficulty of analyzing these textual artifacts of learners in order to offer them feedback represents a serious problem in further extending the usage of chat conversations. PolyCAFe is a system that was designed to support the tutors and to provide automatic feedback for the learners engaged in collaborative chat conversations and discussion forums. The architecture of the system is presented by focusing on two key components: the assessment of the utterances and of the collaborative discourse. PolyCAFe's effectiveness has been proved in a validation experiment with students and tutors from a University course. The main findings from this trial, together with the conclusions of domain experts verifying the accuracy of the assessment provided by PolyCAFe, are also analyzed and commented in detail.

Discourse in instant messenger conversations (chats) with multiple participants is often composed of several intertwining threads. Some chat environments for Computer-Supported Collaborative Learning (CSCL) support and encourage the existence of parallel threads by providing explicit referencing facilities. The paper proposes a discourse model for such chats, based on Mikhail Bakhtin's dialogic theory. It considers that multiple voices (which do not limit to the participants) inter-animate, sometimes in a polyphonic, counterpointal way. An implemented system is also presented, which analyzes such chat logs for detecting additional, implicit links among utterances and threads and, more important for CSCL, for detecting the involvement (inter-animation) of the participants in problem solving. The system begins with a NLP pipe and concludes with inter-animation identification in order to generate feedback and to propose grades for the learners.

Although Computer-Supported Collaborative Learning (CSCL) advocates the use of instant messaging and discussion forums for collaboration between learners, there is a scarcity of tools for leveraging the information in this kind of conversations. Thus, these technologies are primarily used for communication and, once the conversation is over, the raw data is rarely manually analyzed by tutors, teachers and other learners. This paper presents a methodology and a system that can be used for providing feedback and support to learners and tutors that are involved in tasks that make use of chats and forums. In order to achieve this objective, PolyCAFe employs Natural Language Processing and Social Network Analysis techniques to discover polyphony and inter-animation in textual collaborations. To evaluate the proposed approach and the designed system a first validation experiment has been performed and the results are discussed and analyzed in the end of the paper.

The paper describes a complex e-learning experiment that has involved over 700 students that attended the Human-Computer Interaction course at the "Politehnica" University of Bucharest during the last 4 years. The experiment consisted in using social web technologies like blogs and chat conferences to engage students in collaborative learning. The paper presents the learning scenario, the problems encountered and the tools developed for solving these problems and assisting tutors in evaluating the activity of the students. The results of the experiment and of using the blog and chat analysis tools are also covered. Moreover, we show the benefits of using such a scenario for the learning community formed by the students that attended this course in order to supplement the classical teaching and learning paradigm.

Online collaboration among communities of practice using text-based tools, such as instant messaging, forums and web logs (blogs), has become very popular in the last years, but it is difficult to automatically analyze all their content due to the problems of natural language understanding software. However, useful socio-semantic data can be retrieved from a chat conversation using ontology-based text mining techniques. In this paper, a novel approach for detecting several kinds of semantic data from a chat conversation is presented. This method uses a combination of a dialogistic, socio-cultural perspective and of classical knowledge-based text processing methods. Lexical and domain ontologies are used. A tool has been developed for the discovery of the most important topics and of the contribution of each participant in the conversation. The system also discovers new, implicit references among the utterances of the chat in order to offer a multi-voiced representation of the conversation. The application offers a panel for visualizing the threading of the subjects in the chat and the contributions function. The system was experimented on chat sessions of small groups of students participating in courses on Human-Computer Interaction and Natural Language Processing in "Politehnica" University of Bucharest, Romania.

The paper presents an autonomous text classification module for a news web portal for the Romanian language. Statistical natural language processing techniques are combined in order to achieve a completely autonomous functionality of the portal. The news items are automatically collected from a large number of news sources using web syndication. Afterward, machine-learning techniques are used for achieving an automatic classification of the news stream. Firstly, the items are clustered using an agglomerative algorithm and the resulting groups correspond to the main news topics. Thus, more information about each of the main topics is acquired from various news sources. Secondly, text classification algorithms are applied to automatically label each cluster of news items in a predetermined number of classes. More than a thousand news items were employed for both the training and the evaluation of the classifiers. The paper presents a complete comparison of the results obtained for each method.