Aliaksei Severyn

Learning to Rank Short Text Pairs with Convolutional Deep Neural Networks

Learning a similarity function between pairs of objects is at the core of learning to rank approaches. In information retrieval tasks we typically deal with query-document pairs, in question answering -- question-answer pairs. However, before learning can take place, such pairs needs to be mapped from the original space of symbolic words into some feature space encoding various aspects of their relatedness, e.g. lexical, syntactic and semantic. Feature engineering is often a laborious task and may require external knowledge sources that are not always available or difficult to obtain. Recently, deep learning approaches have gained a lot of attention from the research community and industry for their ability to automatically learn optimal feature representation for a given task, while claiming state-of-the-art performance in many tasks in computer vision, speech recognition and natural language processing. In this paper, we present a convolutional neural network architecture for reranking pairs of short texts, where we learn the optimal representation of text pairs and a similarity function to relate them in a supervised way from the available training data. Our network takes only words in the input, thus requiring minimal preprocessing. In particular, we consider the task of reranking short text pairs where elements of the pair are sentences. We test our deep learning system on two popular retrieval tasks from TREC: Question Answering and Microblog Retrieval. Our model demonstrates strong performance on the first task beating previous state-of-the-art systems by about 3\% absolute points in both MAP and MRR and shows comparable results on tweet reranking, while enjoying the benefits of no manual feature engineering and no additional syntactic parsers.

Globally Normalized Transition-Based Neural Networks

We introduce a globally normalized transition-based neural network model that achieves state-of-the-art part-of-speech tagging, dependency parsing and sentence compression results. Our model is a simple feed-forward neural network that operates on a task-specific transition system, yet achieves comparable or better accuracies than recurrent models. We discuss the importance of global as opposed to local normalization: a key insight is that the label bias problem implies that globally normalized models can be strictly more expressive than locally normalized models.

Twitter Sentiment Analysis with Deep Convolutional Neural Networks

This paper describes our deep learning system for sentiment analysis of tweets. The main contribution of this work is a new model for initializing the parameter weights of the convolutional neural network, which is crucial to train an accurate model while avoiding the need to inject any additional features. Briefly, we use an unsupervised neural language model to train initial word embeddings that are further tuned by our deep learning model on a distant supervised corpus. At a final stage, the pre-trained parameters of the network are used to initialize the model. We train the latter on the supervised training data recently made available by the official system evaluation campaign on Twitter Sentiment Analysis organized by Semeval-2015. A comparison between the results of our approach and the systems participating in the challenge on the official test sets, suggests that our model could be ranked in the first two positions in both the phrase-level subtask A (among 11 teams) and on the message-level subtask B (among 40 teams). This is an important evidence on the practical value of our solution.

Structural relationships for large-scale learning of answer re-ranking

Supervised learning applied to answer re-ranking can highly improve on the overall accuracy of question answering (QA) systems. The key aspect is that the relationships and properties of the question/answer pair composed of a question and the supporting passage of an answer candidate, can be efficiently compared with those captured by the learnt model. In this paper, we define novel supervised approaches that exploit structural relationships between a question and their candidate answer passages to learn a re-ranking model. We model structural representations of both questions and answers and their mutual relationships by just using an off-the-shelf shallow syntactic parser. We encode structures in Support Vector Machines (SVMs) by means of sequence and tree kernels, which can implicitly represent question and answer pairs in huge feature spaces. Such models together with the latest approach to fast kernel-based learning enabled the training of our rerankers on hundreds of thousands of instances, which previously rendered intractable for kernelized SVMs. The results on two different QA datasets, e.g., Answerbag and Jeopardy! data, show that our models deliver large improvement on passage re-ranking tasks, reducing the error in Recall of BM25 baseline by about 18%. One of the key findings of this work is that, despite its simplicity, shallow syntactic trees allow for learning complex relational structures, which exhibits a steep learning curve with the increase in the training size.

UNITN: Training Deep Convolutional Neural Network for Twitter Sentiment Classification

This paper describes our deep learning system for sentiment analysis of tweets. The main contribution of this work is a process to initialize the parameter weights of the convolutional neural network, which is crucial to train an accurate model while avoiding the need to inject any additional features. Briefly, we use an unsupervised neural language model to initialize word embeddings that are further tuned by our deep learning model on a distant supervised corpus. At a final stage, the pre-trained parameters of the network are used to initialize the model which is then trained on the supervised training data from Semeval-2015. According to results on the official test sets, our model ranks 1st in the phrase-level subtask A (among 11 teams) and 2nd on the messagelevel subtask B (among 40 teams). Interestingly, computing an average rank over all six test sets (official and five progress test sets) puts our system 1st in both subtasks A and B.

Neural Ranking Models with Weak Supervision

Despite the impressive improvements achieved by unsupervised deep neural networks in computer vision and NLP tasks, such improvements have not yet been observed in ranking for information retrieval. The reason may be the complexity of the ranking problem, as it is not obvious how to learn from queries and documents when no supervised signal is available. Hence, in this paper, we propose to train a neural ranking model using weak supervision, where labels are obtained automatically without human annotators or any external resources (e.g., click data). To this aim, we use the output of an unsupervised ranking model, such as BM25, as a weak supervision signal. We further train a set of simple yet effective ranking models based on feed-forward neural networks. We study their effectiveness under various learning scenarios (point-wise and pair-wise models) and using different input representations (i.e., from encoding query-document pairs into dense/sparse vectors to using word embedding representation). We train our networks using tens of millions of training instances and evaluate it on two standard collections: a homogeneous news collection (Robust) and a heterogeneous large-scale web collection (ClueWeb). Our experiments indicate that employing proper objective functions and letting the networks to learn the input representation based on weakly supervised data leads to impressive performance, with over 13% and 35% MAP improvements over the BM25 model on the Robust and the ClueWeb collections. Our findings also suggest that supervised neural ranking models can greatly benefit from pre-training on large amounts of weakly labeled data that can be easily obtained from unsupervised IR models.

Multi-lingual opinion mining on YouTube

We designed the first model for effectively carrying out opinion mining on YouTube comments.We propose kernel methods applied to a robust shallow syntactic structure, which improves accuracy for both languages.Our approach greatly outperforms other basic models on cross-domain settings.We created a YouTube corpus (in Italian and English) and made it available for the research community.Comments must be classified in subcategories to make opinion mining effective on YouTube. In order to successfully apply opinion mining (OM) to the large amounts of user-generated content produced every day, we need robust models that can handle the noisy input well yet can easily be adapted to a new domain or language. We here focus on opinion mining for YouTube by (i) modeling classifiers that predict the type of a comment and its polarity, while distinguishing whether the polarity is directed towards the product or video; (ii) proposing a robust shallow syntactic structure (STRUCT) that adapts well when tested across domains; and (iii) evaluating the effectiveness on the proposed structure on two languages, English and Italian. We rely on tree kernels to automatically extract and learn features with better generalization power than traditionally used bag-of-word models. Our extensive empirical evaluation shows that (i) STRUCT outperforms the bag-of-words model both within the same domain (up to 2.6% and 3% of absolute improvement for Italian and English, respectively); (ii) it is particularly useful when tested across domains (up to more than 4% absolute improvement for both languages), especially when little training data is available (up to 10% absolute improvement) and (iii) the proposed structure is also effective in a lower-resource language scenario, where only less accurate linguistic processing tools are available.

Large-scale support vector learning with structural kernels

In this paper, we present an extensive study of the cutting-plane algorithm (CPA) applied to structural kernels for advanced text classification on large datasets. In particular, we carry out a comprehensive experimentation on two interesting natural language tasks, e.g. predicate argument extraction and question answering. Our results show that (i) CPA applied to train a non-linear model with different tree kernels fully matches the accuracy of the conventional SVM algorithm while being ten times faster; (ii) by using smaller sampling sizes to approximate subgradients in CPA we can trade off accuracy for speed, yet the optimal parameters and kernels found remain optimal for the exact SVM. These results open numerous research perspectives, e.g. in natural language processing, as they show that complex structural kernels can be efficiently used in real-world applications. For example, for the first time, we could carry out extensive tests of several tree kernels on millions of training instances. As a direct benefit, we could experiment with a variant of the partial tree kernel, which we also propose in this paper.

Leveraging Large Amounts of Weakly Supervised Data for Multi-Language Sentiment Classification

This paper presents a novel approach for multi-lingual sentiment classification in short texts. This is a challenging task as the amount of training data in languages other than English is very limited. Previously proposed multi-lingual approaches typically require to establish a correspondence to English for which powerful classifiers are already available. In contrast, our method does not require such supervision. We leverage large amounts of weakly-supervised data in various languages to train a multi-layer convolutional network and demonstrate the importance of using pre-training of such networks. We thoroughly evaluate our approach on various multi-lingual datasets, including the recent SemEval-2016 sentiment prediction benchmark (Task 4), where we achieved state-of-the-art performance. We also compare the performance of our model trained individually for each language to a variant trained for all languages at once. We show that the latter model reaches slightly worse - but still acceptable - performance when compared to the single language model, while benefiting from better generalization properties across languages.

On the Automatic Learning of Sentiment Lexicons

This paper describes a simple and principled approach to automatically construct sentiment lexicons using distant supervision. We induce the sentiment association scores for the lexicon items from a model trained on a weakly supervised corpora. Our empirical findings show that features extracted from such a machine-learned lexicon outperform models using manual or other automatically constructed sentiment lexicons. Finally, our system achieves the state-of-the-art in Twitter Sentiment Analysis tasks from Semeval-2013 and ranks 2nd best in Semeval-2014 according to the average rank.