Funnelling: A New Ensemble Method for Heterogeneous Transfer Learning and its Application to Cross-Lingual Text Classification

Cross-lingual Text Classification (CLC) consists of automatically classifying, according to a common set C of classes, documents each written in one of a set of languages L, and doing so more accurately than when naively classifying each document via its corresponding language-specific classifier. In order to obtain an increase in the classification accuracy for a given language, the system thus needs to also leverage the training examples written in the other languages. We tackle multilabel CLC via funnelling, a new ensemble learning method that we propose here. Funnelling consists of generating a two-tier classification system where all documents, irrespectively of language, are classified by the same (2nd-tier) classifier. For this classifier all documents are represented in a common, language-independent feature space consisting of the posterior probabilities generated by 1st-tier, language-dependent classifiers. This allows the classification of all test documents, of any language, to benefit from the information present in all training documents, of any language. We present substantial experiments, run on publicly available multilingual text collections, in which funnelling is shown to significantly outperform a number of state-of-the-art baselines. All code and datasets (in vector form) are made publicly available.

Building Automated Survey Coders via Interactive Machine Learning

Software systems trained via machine learning to automatically classify open-ended answers (a.k.a. verbatims) are by now a reality. Still, their adoption in the survey coding industry has been less widespread than it might have been. Among the factors that have hindered a more massive takeup of this technology are the effort involved in manually coding a sufficient amount of training data, the fact that small studies do not seem to justify this effort, and the fact that the process needs to be repeated anew when brand new coding tasks arise. In this paper we will argue for an approach to building verbatim classifiers that we will call "Interactive Learning", and that addresses all the above problems. We will show that, for the same amount of training effort, interactive learning delivers much better coding accuracy than standard "non-interactive" learning. This is especially true when the amount of data we are willing to manually code is small, which makes this approach attractive also for small-scale studies. Interactive learning also lends itself to reusing previously trained classifiers for dealing with new (albeit related) coding tasks. Interactive learning also integrates better in the daily workflow of the survey specialist, and delivers a better user experience overall.

Learning to Weight for Text Classification

In information retrieval (IR) and related tasks, term weighting approaches typically consider the frequency of the term in the document and in the collection in order to compute a score reflecting the importance of the term for the document. In tasks characterized by the presence of training data (such as text classification) it seems logical that the term weighting function should take into account the distribution (as estimated from training data) of the term across the classes of interest. Although `supervised term weighting' approaches that use this intuition have been described before, they have failed to show consistent improvements. In this article we analyse the possible reasons for this failure, and call consolidated assumptions into question. Following this criticism we propose a novel supervised term weighting approach that, instead of relying on any predefined formula, learns a term weighting function optimised on the training set of interest; we dub this approach \emph{Learning to Weight} (LTW). The experiments that we run on several well-known benchmarks, and using different learning methods, show that our method outperforms previous term weighting approaches in text classification.

Revisiting Distributional Correspondence Indexing: A Python Reimplementation and New Experiments

This paper introduces PyDCI, a new implementation of Distributional Correspondence Indexing (DCI) written in Python. DCI is a transfer learning method for cross-domain and cross-lingual text classification for which we had provided an implementation (here called JaDCI) built on top of JaTeCS, a Java framework for text classification. PyDCI is a stand-alone version of DCI that exploits scikit-learn and the SciPy stack. We here report on new experiments that we have carried out in order to test PyDCI, and in which we use as baselines new high-performing methods that have appeared after DCI was originally proposed. These experiments show that, thanks to a few subtle ways in which we have improved DCI, PyDCI outperforms both JaDCI and the above-mentioned high-performing methods, and delivers the best known results on the two popular benchmarks on which we had tested DCI, i.e., MultiDomainSentiment (a.k.a. MDS -- for cross-domain adaptation) and Webis-CLS-10 (for cross-lingual adaptation). PyDCI, together with the code allowing to replicate our experiments, is available at https://github.com/AlexMoreo/pydci .

Evaluation Measures for Quantification: An Axiomatic Approach

Quantification is the task of estimating, given a set $\sigma$ of unlabelled items and a set of classes $\mathcal{C}=\{c_{1}, \ldots, c_{|\mathcal{C}|}\}$, the prevalence (or `relative frequency') in $\sigma$ of each class $c_{i}\in \mathcal{C}$. While quantification may in principle be solved by classifying each item in $\sigma$ and counting how many such items have been labelled with $c_{i}$, it has long been shown that this `classify and count' (CC) method yields suboptimal quantification accuracy. As a result, quantification is no longer considered a mere byproduct of classification, and has evolved as a task of its own. While the scientific community has devoted a lot of attention to devising more accurate quantification methods, it has not devoted much to discussing what properties an \emph{evaluation measure for quantification} (EMQ) should enjoy, and which EMQs should be adopted as a result. This paper lies down a number of interesting properties that an EMQ may or may not enjoy, discusses if (and when) each of these properties is desirable, surveys the EMQs that have been used so far, and discusses whether they enjoy or not the above properties. As a result of this investigation, some of the EMQs that have been used in the literature turn out to be severely unfit, while others emerge as closer to what the quantification community actually needs. However, a significant result is that no existing EMQ satisfies all the properties identified as desirable, thus indicating that more research is needed in order to identify (or synthesize) a truly adequate EMQ.

A Recurrent Neural Network for Sentiment Quantification

Quantification is a supervised learning task that consists in predicting, given a set of classes C and a set D of unlabelled items, the prevalence (or relative frequency) p(c|D) of each class c in C. Quantification can in principle be solved by classifying all the unlabelled items and counting how many of them have been attributed to each class. However, this "classify and count" approach has been shown to yield suboptimal quantification accuracy; this has established quantification as a task of its own, and given rise to a number of methods specifically devised for it. We propose a recurrent neural network architecture for quantification (that we call QuaNet) that observes the classification predictions to learn higher-order "quantification embeddings", which are then refined by incorporating quantification predictions of simple classify-and-count-like methods. We test {QuaNet on sentiment quantification on text, showing that it substantially outperforms several state-of-the-art baselines.

Optimizing Non-decomposable Measures with Deep Networks

We present a class of algorithms capable of directly training deep neural networks with respect to large families of task-specific performance measures such as the F-measure and the Kullback-Leibler divergence that are structured and non-decomposable. This presents a departure from standard deep learning techniques that typically use squared or cross-entropy loss functions (that are decomposable) to train neural networks. We demonstrate that directly training with task-specific loss functions yields much faster and more stable convergence across problems and datasets. Our proposed algorithms and implementations have several novel features including (i) convergence to first order stationary points despite optimizing complex objective functions; (ii) use of fewer training samples to achieve a desired level of convergence, (iii) a substantial reduction in training time, and (iv) a seamless integration of our implementation into existing symbolic gradient frameworks. We implement our techniques on a variety of deep architectures including multi-layer perceptrons and recurrent neural networks and show that on a variety of benchmark and real data sets, our algorithms outperform traditional approaches to training deep networks, as well as some recent approaches to task-specific training of neural networks.

Online Optimization Methods for the Quantification Problem

The estimation of class prevalence, i.e., the fraction of a population that belongs to a certain class, is a very useful tool in data analytics and learning, and finds applications in many domains such as sentiment analysis, epidemiology, etc. For example, in sentiment analysis, the objective is often not to estimate whether a specific text conveys a positive or a negative sentiment, but rather estimate the overall distribution of positive and negative sentiments during an event window. A popular way of performing the above task, often dubbed quantification, is to use supervised learning to train a prevalence estimator from labeled data. Contemporary literature cites several performance measures used to measure the success of such prevalence estimators. In this paper we propose the first online stochastic algorithms for directly optimizing these quantification-specific performance measures. We also provide algorithms that optimize hybrid performance measures that seek to balance quantification and classification performance. Our algorithms present a significant advancement in the theory of multivariate optimization and we show, by a rigorous theoretical analysis, that they exhibit optimal convergence. We also report extensive experiments on benchmark and real data sets which demonstrate that our methods significantly outperform existing optimization techniques used for these performance measures.

Optimizing Text Quantifiers for Multivariate Loss Functions

We address the problem of \emph{quantification}, a supervised learning task whose goal is, given a class, to estimate the relative frequency (or \emph{prevalence}) of the class in a dataset of unlabelled items. Quantification has several applications in data and text mining, such as estimating the prevalence of positive reviews in a set of reviews of a given product, or estimating the prevalence of a given support issue in a dataset of transcripts of phone calls to tech support. So far, quantification has been addressed by learning a general-purpose classifier, counting the unlabelled items which have been assigned the class, and tuning the obtained counts according to some heuristics. In this paper we depart from the tradition of using general-purpose classifiers, and use instead a supervised learning model for \emph{structured prediction}, capable of generating classifiers directly optimized for the (multivariate and non-linear) function used for evaluating quantification accuracy. The experiments that we have run on 5500 binary high-dimensional datasets (averaging more than 14,000 documents each) show that this method is more accurate, more stable, and more efficient than existing, state-of-the-art quantification methods.

On the Effects of Low-Quality Training Data on Information Extraction from Clinical Reports

In the last five years there has been a flurry of work on information extraction from clinical documents, i.e., on algorithms capable of extracting, from the informal and unstructured texts that are generated during everyday clinical practice, mentions of concepts relevant to such practice. Most of this literature is about methods based on supervised learning, i.e., methods for training an information extraction system from manually annotated examples. While a lot of work has been devoted to devising learning methods that generate more and more accurate information extractors, no work has been devoted to investigating the effect of the quality of training data on the learning process. Low quality in training data often derives from the fact that the person who has annotated the data is different from the one against whose judgment the automatically annotated data must be evaluated. In this paper we test the impact of such data quality issues on the accuracy of information extraction systems as applied to the clinical domain. We do this by comparing the accuracy deriving from training data annotated by the authoritative coder (i.e., the one who has also annotated the test data, and by whose judgment we must abide), with the accuracy deriving from training data annotated by a different coder. The results indicate that, although the disagreement between the two coders (as measured on the training set) is substantial, the difference is (surprisingly enough) not always statistically significant.