Search

Research into the nature and structure of ‘Dark Webs’ such as Tor has largely focused upon manually labelling a series of crawled sites against a series of categories, sometimes using these labels as a training corpus for subsequent automated crawls. Such an approach is adequate for establishing broad taxonomies, but is of limited value for specialised tasks within the field of law enforcement. Contrastingly, existing research into illicit behaviour online has tended to focus upon particular crime types such as terrorism. A gap exists between taxonomies capable of holistic representation and those capable of detailing criminal behaviour. The absence of such a taxonomy limits interoperability between agencies, curtailing development of standardised classification tools. We introduce the Tor-use Motivation Model (TMM), a two-dimensional classification methodology specifically designed for use within a law enforcement context. The TMM achieves greater levels of granularity by explicitly distinguishing site content from motivation, providing a richer labelling schema without introducing inefficient complexity or reliance upon overly broad categories of relevance. We demonstrate this flexibility and robustness through direct examples, showing the TMM's ability to distinguish a range of unethical and illegal behaviour without bloating the model with unnecessary detail. The authors of this paper received permission from the Australian government to conduct an unrestricted crawl of Tor for research purposes, including the gathering and analysis of illegal materials such as child pornography. The crawl gathered 232,792 pages from 7651 Tor virtual domains, resulting in the collation of a wide spectrum of materials, from illicit to downright banal. Existing conceptual models and their labelling schemas were tested against a small sample of gathered data, and were observed to be either overly prescriptive or vague for law enforcement purposes - particularly when used for prioritising sites of interest for further investigation. In this paper we deploy the TMM by manually labelling a corpus of over 4000 unique Tor pages. We found a network impacted (but not dominated) by illicit commerce and money laundering, but almost completely devoid of violence and extremism. In short, criminality on this ‘dark web’ is based more upon greed and desire, rather than any particular political motivations.

Extracting structured information from unstructured text is important for the qualitative data analysis. Leveraging NLP techniques for qualitative data analysis will effectively accelerate the annotation process, allow for large-scale analysis and provide more insights into the text to improve the performance. The first step for gaining insights from the text is Named Entity Recognition (NER). A significant challenge that directly impacts the performance of the NER process is the domain diversity in qualitative data. The represented text varies according to its domain in many aspects including taxonomies, length, formality and format. In this paper we discuss and analyse the performance of state-of-the-art tools across domains to elaborate their robustness and reliability. In order to do that, we developed a standard, expandable and flexible framework to analyse and test tools performance using corpora representing text across various domains. We performed extensive analysis and comparison of tools across various domains and from various perspectives. The resulting comparison and analysis are of significant importance for providing a holistic illustration of the state-of-the-art tools.

Recent advances have demonstrated substantial benefits from learning with both generative and discriminative parameters. On the one hand, generative approaches address the estimation of the parameters of the joint distribution--$$\mathrm{P}(y,\mathbf{x})$$P(y,x), which for most network types is very computationally efficient (a notable exception to this are Markov networks) and on the other hand, discriminative approaches address the estimation of the parameters of the posterior distribution--and, are more effective for classification, since they fit $$\mathrm{P}(y|\mathbf{x})$$P(y|x) directly. However, discriminative approaches are less computationally efficient as the normalization factor in the conditional log-likelihood precludes the derivation of closed-form estimation of parameters. This paper introduces a new discriminative parameter learning method for Bayesian network classifiers that combines in an elegant fashion parameters learned using both generative and discriminative methods. The proposed method is discriminative in nature, but uses estimates of generative probabilities to speed-up the optimization process. A?second contribution is to propose a simple framework to characterize the parameter learning task for Bayesian network classifiers. We conduct an extensive set of experiments on 72 standard datasets and demonstrate that our proposed discriminative parameterization provides an efficient alternative to other state-of-the-art parameterizations.

Density-based clustering algorithms are able to identify clusters of arbitrary shapes and sizes in a dataset which contains noise. It is well-known that most of these algorithms, which use a global density threshold, have difficulty identifying all clusters in a dataset having clusters of greatly varying densities. This paper identifies and analyses the condition under which density-based clustering algorithms fail in this scenario. It proposes a density-ratio based method to overcome this weakness, and reveals that it can be implemented in two approaches. One approach is to modify a density-based clustering algorithm to do density-ratio based clustering by using its density estimator to compute density-ratio. The other approach involves rescaling the given dataset only. An existing density-based clustering algorithm, which is applied to the rescaled dataset, can find all clusters with varying densities that would otherwise impossible had the same algorithm been applied to the unscaled dataset. We provide an empirical evaluation using DBSCAN, OPTICS and SNN to show the effectiveness of these two approaches. HighlightsAnalyse a key weakness of density-based clustering algorithms.Introduce two approaches based on density-ratio to overcome this weakness.ReCon converts an existing density estimator to a density-ratio estimator.ReScale transforms a dataset by an adaptive scaling based on density-ratio.ReCon and ReScale approaches improve three density-based clustering algorithms.

Current random-forest (RF)-based learning-to-rank (LtR) algorithms use a classification or regression framework to solve the ranking problem in a pointwise manner. The success of this simple yet effective approach coupled with the inherent parallelizability of the learning algorithm makes it a strong candidate for widespread adoption. In this article, we aim to better understand the effectiveness of RF-based rank-learning algorithms with a focus on the comparison between pointwise and listwise approaches. We introduce what we believe to be the first listwise version of an RF-based LtR algorithm. The algorithm directly optimizes an information retrieval metric of choice (in our case, NDCG) in a greedy manner. Direct optimization of the listwise objective functions is computationally prohibitive for most learning algorithms, but possible in RF since each tree maximizes the objective in a coordinate-wise fashion. Computational complexity of the listwise approach is higher than the pointwise counterpart; hence for larger datasets, we design a hybrid algorithm that combines a listwise objective in the early stages of tree construction and a pointwise objective in the latter stages. We also study the effect of the discount function of NDCG on the listwise algorithm. Experimental results on several publicly available LtR datasets reveal that the listwise/hybrid algorithm outperforms the pointwise approach on the majority (but not all) of the datasets. We then investigate several aspects of the two algorithms to better understand the inevitable performance tradeoffs. The aspects include examining an RF-based unsupervised LtR algorithm and comparing individual tree strength. Finally, we compare the the investigated RF-based algorithms with several other LtR algorithms.

This paper introduces Accelerated Logistic Regression: a hybrid generative-discriminative approach to training Logistic Regression with high-order features. We present two main results: (1) that our combined generative-discriminative approach significantly improves the efficiency of Logistic Regression and (2) that incorporating higher order features (i.e. features that are the Cartesian products of the original features) reduces the bias of Logistic Regression, which in turn significantly reduces its error on large datasets. We assess the efficacy of Accelerated Logistic Regression by conducting an extensive set of experiments on 75 standard datasets. We demonstrate its competitiveness, particularly on large datasets, by comparing against state-of-the-art classifiers including Random Forest and Averaged n-Dependence Estimators.

This paper investigates the annotation-based retrieval (AR) of World Wide Web (WWW) resources that has been annotated by users on Collaborative Tagging (CT) platforms as a form of user-generated content (UGC). Previous approaches have simply weight the WWW resources according to their popularity, in order to leverage on the inherent wisdom of the crowd (WotC). In this paper, we argue that the popularity alone is not a sufficient indicator of quality since (1) some users are better annotators than the others; (2) resource popularity can be easily inflated by malicious users; and (3) high quality but highly specific resources may exhibit lower popularity than more general ones. Thus, we investigate the indicators of information quality for WWW resources, particularly user annotations that can be used to describe them. This research estimates the user expertise of content annotators in order to infer the information quality of their contributions; by exploring the various signals available on social bookmarking platforms such as the temporal information of annotations. The evaluation in retrieval performance on social bookmarking data shows significant improvements with the estimated user expertise and inferred information quality.

Modern Knowledge Graphs such as DBPedia contain significant information regarding Named Entities and the logical relationships which exist between them. Twitter on the other hand, contains important information on the popularity and frequency with which these entities are mentioned and discussed in combination with one another. In this paper we investigate whether these two sources of information can be used to complement and explain one another. In particular, we would like to know whether the logical relationships (a.k.a. semantic paths) which exist between pairs of known entities can help to explain the frequency with which those entities co-occur with one another in Twitter. To do this we train a ranking function over semantic paths between pairs of entities. The aim of the ranker is to identify the path that most likely explains why a particular pair of entities have appeared together in a particular tweet. We train the ranking model using a number of lexical, graph-embedding and popularity-based features over semantic paths containing a single intermediate entity and demonstrate the efficacy of the model for determining why pairs of entities occur together in tweets.

Reddit as a social curation site relies on its users to curate content from the World Wide Web (WWW) for the consumption of other users. Content on the site is enriched through user comments, discussions and extensions. This additional content is of varying quality however -- ranging from meaningful information to misleading content; depending on the reliability, expertise and intention of the authors. Reddit relies on the Wisdom of the Crowd (WotC) from its community as well as selected moderators to manage its content. We argue that this approach suffers from the cold start in collecting user votes and is at risk of user bias, particularly a group-think mentality. Besides that, managing the large collection of content on Reddit is expensive. In our study, we explore the estimation of relative user expertise through various content-agnostic approaches. We show that it is possible to infer information quality on Reddit using the expertise of the authors. This prediction of content quality could lead to an improved organisation of Reddit content (re-ranking) for user consumption and future information retrieval.

Comparing NLP systems to select the best one for a task of interest, such as named entity recognition, is critical for practitioners and researchers. A rigorous approach involves setting up a hypothesis testing scenario using the performance of the systems on query documents. However, often the hypothesis testing approach needs to send a lot of document queries to the systems, which can be problematic. In this paper, we present an effective alternative based on the multi-armed bandit (MAB). We propose a hierarchical generative model to represent the uncertainty in the performance measures of the competing systems, to be used by Thompson Sampling to solve the resulting MAB. Experimental results on both synthetic and real data show that our approach requires significantly fewer queries compared to the standard benchmarking technique to identify the best system according to F-measure.

Community Question-Answering (CQA) platforms leverage the inherent wisdom of the crowd - enabling users to retrieve quality information from domain experts through natural language. An important and challenging task is to identify reliable and trusted experts on large popular CQA platforms. State-of-the-art graph-based approaches to expertise estimation consider only user-user interactions without taking the relative contribution of individual answers into account, while pairwise-comparison approaches consider only pairs involving the best-answerer of each question. This research argues that there is a need to account for the user's relative contribution towards solving the question when estimating user expertise and proposes a content-agnostic measure of user contributions. This addition is incorporated into a competition-based approach for ranking users' question answering ability. The paper analyses how improvements in user expertise estimation impact on applications in expert search and answer quality prediction. Experiments using the Yahoo! Chiebukuro data show encouraging performance improvements and robustness over state-of-the-art approaches.

We study the problem of generating DAG-structured category hierarchies over a given set of documents associated with "importance" scores. Example application includes automatically generating Wikipedia disambiguation pages for a set of articles having click counts associated with them. Unlike previous works, which focus on clustering the set of documents using the category hierarchy as features, we directly pose the problem as that of finding a DAG structured generative mode that has maximum likelihood of generating the observed "importance" scores for each document where documents are modeled as the leaf nodes in the DAG structure. Desirable properties of the categories in the inferred DAG-structured hierarchy include document coverage and category relevance, each of which, we show, is naturally modeled by our generative model. We propose two different algorithms for estimating the model parameters. One by modeling the DAG as a Bayesian Network and estimating its parameters via Gibbs Sampling; and the other by estimating the path probabilities using the Expectation Maximization algorithm. We empirically evaluate our method on the problem of automatically generating Wikipedia disambiguation pages using human generated clusterings as the ground truth. We find that our framework improves upon the baselines according to the F1 score and Entropy that are used as standard metrics to evaluate the hierarchical clustering.

Query-level instance weighting is a technique for unsupervised transfer ranking, which aims to train a ranker on a source collection so that it also performs effectively on a target collection, even if no judgement information exists for the latter. Past work has shown that this approach can be used to significantly improve effectiveness; in this work, the approach is re-examined on a wide set of publicly available L2R test collections with more advanced learning to rank algorithms. Different query-level weighting strategies are examined against two transfer ranking frameworks: AdaRank and a new weighted LambdaMART algorithm. Our experimental results show that the effectiveness of different weighting strategies, including those shown in past work, vary under different transferring environments. In particular, (i) Kullback-Leibler based density-ratio estimation tends to outperform a classification-based approach and (ii) aggregating document-level weights into query-level weights is likely superior to direct estimation using a query-level representation. The Nemenyi statistical test, applied across multiple datasets, indicates that most weighting transfer learning methods do not significantly outperform baselines, although there is potential for the further development of such techniques.

This paper introduces the first generic version of data dependent dissimilarity and shows that it provides a better closest match than distance measures for three existing algorithms in clustering, anomaly detection and multi-label classification. For each algorithm, we show that by simply replacing the distance measure with the data dependent dissimilarity measure, it overcomes a key weakness of the otherwise unchanged algorithm.

Automatic sarcasm detection is the task of predicting sarcasm in text. This is a crucial step to sentiment analysis, considering prevalence and challenges of sarcasm in sentiment-bearing text. Beginning with an approach that used speech-based features, sarcasm detection has witnessed great interest from the sentiment analysis community. This paper is the first known compilation of past work in automatic sarcasm detection. We observe three milestones in the research so far: semi-supervised pattern extraction to identify implicit sentiment, use of hashtag-based supervision, and use of context beyond target text. In this paper, we describe datasets, approaches, trends and issues in sarcasm detection. We also discuss representative performance values, shared tasks and pointers to future work, as given in prior works. In terms of resources that could be useful for understanding state-of-the-art, the survey presents several useful illustrations - most prominently, a table that summarizes past papers along different dimensions such as features, annotation techniques, data forms, etc., This paper is likely to be submitted to ACM CSUR. This copy on arXiv is to obtain feedback from stakeholders

This paper is a novel study that views sarcasm detection in dialogue as a sequence labeling task, where a dialogue is made up of a sequence of utterances. We create a manuallylabeled dataset of dialogue from TV series ‘Friends’ annotated with sarcasm. Our goal is to predict sarcasm in each utterance, using sequential nature of a scene. We show performance gain using sequence labeling as compared to classification-based approaches. Our experiments are based on three sets of features, one is derived from information in our dataset, the other two are from past works. Two sequence labeling algorithms (SVM-HMM and SEARN) outperform three classification algorithms (SVM, Naive Bayes) for all these feature sets, with an increase in F-score of around 4%. Our observations highlight the viability of sequence labeling techniques for sarcasm detection of dialogue.

People often use social media to discuss opinions, including political ones. We refer to relevant topics in these discussions as political issues, and the alternate stands towards these topics as political positions. We present a Political Issue Extraction (PIE) model that is capable of discovering political issues and positions from an unlabeled dataset of tweets. A strength of this model is that it uses twitter timelines of political and non-political authors, and affiliation information of only political authors. The model estimates word-specific distributions (that denote political issues and positions) and hierarchical author/group-specific distributions (that show how these issues divide people). Our experiments using a dataset of 2.4 million tweets from the US show that this model effectively captures the desired properties (with respect to words and groups) of political discussions. We also evaluate the two components of the model by experimenting with: (a) Use to alternate strategies to classify words, and (b) Value addition due to incorporation of group membership information. Estimated distributions are then used to predict political affiliation with 68% accuracy.