Erion Çano

Amir Ziaee, Erion Çano

This study introduces a new normalization layer termed Batch Layer Normalization (BLN) to reduce the problem of internal covariate shift in deep neural network layers. As a combined version of batch and layer normalization, BLN adaptively puts appropriate weight on mini-batch and feature normalization based on the inverse size of mini-batches to normalize the input to a layer during the learning process. It also performs the exact computation with a minor change at inference times, using either mini-batch statistics or population statistics. The decision process to either use statistics of mini-batch or population gives BLN the ability to play a comprehensive role in the hyper-parameter optimization process of models. The key advantage of BLN is the support of the theoretical analysis of being independent of the input data, and its statistical configuration heavily depends on the task performed, the amount of training data, and the size of batches. Test results indicate the application potential of BLN and its faster convergence than batch normalization and layer normalization in both Convolutional and Recurrent Neural Networks. The code of the experiments is publicly available online (https://github.com/A2Amir/Batch-Layer-Normalization).

Stefan Schweter, Luisa März, Katharina Schmid et al.

Compared to standard Named Entity Recognition (NER), identifying persons, locations, and organizations in historical texts constitutes a big challenge. To obtain machine-readable corpora, the historical text is usually scanned and Optical Character Recognition (OCR) needs to be performed. As a result, the historical corpora contain errors. Also, entities like location or organization can change over time, which poses another challenge. Overall, historical texts come with several peculiarities that differ greatly from modern texts and large labeled corpora for training a neural tagger are hardly available for this domain. In this work, we tackle NER for historical German, English, French, Swedish, and Finnish by training large historical language models. We circumvent the need for large amounts of labeled data by using unlabeled data for pretraining a language model. We propose hmBERT, a historical multilingual BERT-based language model, and release the model in several versions of different sizes. Furthermore, we evaluate the capability of hmBERT by solving downstream NER as part of this year's HIPE-2022 shared task and provide detailed analysis and insights. For the Multilingual Classical Commentary coarse-grained NER challenge, our tagger HISTeria outperforms the other teams' models for two out of three languages.

Erion Çano, Benjamin Roth

Collections of research article data harvested from the web have become common recently since they are important resources for experimenting on tasks such as named entity recognition, text summarization, or keyword generation. In fact, certain types of experiments require collections that are both large and topically structured, with records assigned to separate research disciplines. Unfortunately, the current collections of publicly available research articles are either small or heterogeneous and unstructured. In this work, we perform topic segmentation of a paper data collection that we crawled and produce a multitopic dataset of roughly seven million paper data records. We construct a taxonomy of topics extracted from the data records and then annotate each document with its corresponding topic from that taxonomy. As a result, it is possible to use this newly proposed dataset in two modalities: as a heterogeneous collection of documents from various disciplines or as a set of homogeneous collections, each from a single research topic.

Erion Çano

Lack of available resources such as text corpora for low-resource languages seriously hinders research on natural language processing and computational linguistics. This paper presents AlbMoRe, a corpus of 800 sentiment annotated movie reviews in Albanian. Each text is labeled as positive or negative and can be used for sentiment analysis research. Preliminary results based on traditional machine learning classifiers trained with the AlbMoRe samples are also reported. They can serve as comparison baselines for future research experiments.

Erion Çano

Scarcity of resources such as annotated text corpora for under-resourced languages like Albanian is a serious impediment in computational linguistics and natural language processing research. This paper presents AlbNER, a corpus of 900 sentences with labeled named entities, collected from Albanian Wikipedia articles. Preliminary results with BERT and RoBERTa variants fine-tuned and tested with AlbNER data indicate that model size has slight impact on NER performance, whereas language transfer has a significant one. AlbNER corpus and these obtained results should serve as baselines for future experiments.

Mousumi Akter, Erion Çano, Erik Weber et al.

This article provides a systematic up-to-date survey of automatic summarization techniques, datasets, models, and evaluation methods in the legal domain. Through specific source selection criteria, we thoroughly review over 120 papers spanning the modern `transformer' era of natural language processing (NLP), thus filling a gap in existing systematic surveys on the matter. We present existing research along several axes and discuss trends, challenges, and opportunities for future research.

Anna Leschanowsky, Zahra Kolagar, Erion Çano et al.

The transparency principle of the General Data Protection Regulation (GDPR) requires data processing information to be clear, precise, and accessible. While language models show promise in this context, their probabilistic nature complicates truthfulness and comprehensibility. This paper examines state-of-the-art Retrieval Augmented Generation (RAG) systems enhanced with alignment techniques to fulfill GDPR obligations. We evaluate RAG systems incorporating an alignment module like Rewindable Auto-regressive Inference (RAIN) and our proposed multidimensional extension, MultiRAIN, using a Privacy Q&A dataset. Responses are optimized for preciseness and comprehensibility and are assessed through 21 metrics, including deterministic and large language model-based evaluations. Our results show that RAG systems with an alignment module outperform baseline RAG systems on most metrics, though none fully match human answers. Principal component analysis of the results reveals complex interactions between metrics, highlighting the need to refine metrics. This study provides a foundation for integrating advanced natural language processing systems into legal compliance frameworks.

Erion Çano, Dario Lamaj

The scarcity of available text corpora for low-resource languages like Albanian is a serious hurdle for research in natural language processing tasks. This paper introduces AlbNews, a collection of 600 topically labeled news headlines and 2600 unlabeled ones in Albanian. The data can be freely used for conducting topic modeling research. We report the initial classification scores of some traditional machine learning classifiers trained with the AlbNews samples. These results show that basic models outrun the ensemble learning ones and can serve as a baseline for future experiments.

Erion Çano, Ivan Habernal

Ensuring user privacy by synthesizing data from large language models (LLMs) tuned under differential privacy (DP) has become popular recently. However, the impact of DP fine-tuned LLMs on the quality of the language and the utility of the texts they produce has not been investigated. In this work, we tune five LLMs with three corpora under four levels of privacy and assess the length, the grammatical correctness, and the lexical diversity of the text outputs they produce. We also probe the utility of the synthetic outputs in downstream classification tasks such as book genre recognition based on book descriptions and cause of death recognition based on verbal autopsies. The results indicate that LLMs tuned under stronger privacy constrains produce texts that are shorter by at least 77 %, that are less grammatically correct by at least 9 %, and are less diverse by at least 10 % in bi-gram diversity. Furthermore, the accuracy they reach in downstream classification tasks decreases, which might be detrimental to the usefulness of the generated synthetic data.

Vasiliki Kougia, Simon Fetzel, Thomas Kirchmair et al.

Memes are a popular form of communicating trends and ideas in social media and on the internet in general, combining the modalities of images and text. They can express humor and sarcasm but can also have offensive content. Analyzing and classifying memes automatically is challenging since their interpretation relies on the understanding of visual elements, language, and background knowledge. Thus, it is important to meaningfully represent these sources and the interaction between them in order to classify a meme as a whole. In this work, we propose to use scene graphs, that express images in terms of objects and their visual relations, and knowledge graphs as structured representations for meme classification with a Transformer-based architecture. We compare our approach with ImgBERT, a multimodal model that uses only learned (instead of structured) representations of the meme, and observe consistent improvements. We further provide a dataset with human graph annotations that we compare to automatically generated graphs and entity linking. Analysis shows that automatic methods link more entities than human annotators and that automatically generated graphs are better suited for hatefulness classification in memes.

Benjamin Roth, Erion Çano

We propose a scheme for self-training of grammaticality models for constituency analysis based on linguistic tests. A pre-trained language model is fine-tuned by contrastive estimation of grammatical sentences from a corpus, and ungrammatical sentences that were perturbed by a syntactic test, a transformation that is motivated by constituency theory. We show that consistent gains can be achieved if only certain positive instances are chosen for training, depending on whether they could be the result of a test transformation. This way, the positives, and negatives exhibit similar characteristics, which makes the objective more challenging for the language model, and also allows for additional markup that indicates the position of the test application within the sentence.

Erion Çano, Ondřej Bojar

Being able to predict the length of a scientific paper may be helpful in numerous situations. This work defines the paper length prediction task as a regression problem and reports several experimental results using popular machine learning models. We also create a huge dataset of publication metadata and the respective lengths in number of pages. The dataset will be freely available and is intended to foster research in this domain. As future work, we would like to explore more advanced regressors based on neural networks and big pretrained language models.

Erion Çano, Maurizio Morisio

The fabulous results of convolution neural networks in image-related tasks, attracted attention of text mining, sentiment analysis and other text analysis researchers. It is however difficult to find enough data for feeding such networks, optimize their parameters, and make the right design choices when constructing network architectures. In this paper we present the creation steps of two big datasets of song emotions. We also explore usage of convolution and max-pooling neural layers on song lyrics, product and movie review text datasets. Three variants of a simple and flexible neural network architecture are also compared. Our intention was to spot any important patterns that can serve as guidelines for parameter optimization of similar models. We also wanted to identify architecture design choices which lead to high performing sentiment analysis models. To this end, we conducted a series of experiments with neural architectures of various configurations. Our results indicate that parallel convolutions of filter lengths up to three are usually enough for capturing relevant text features. Also, max-pooling region size should be adapted to the length of text documents for producing the best feature maps. Top results we got are obtained with feature maps of lengths 6 to 18. An improvement on future neural network models for sentiment analysis, could be generating sentiment polarity prediction of documents using aggregation of predictions on smaller excerpt of the entire text.

Erion Çano, Riccardo Coppola, Eleonora Gargiulo et al.

Driving and music listening are two inseparable everyday activities for millions of people today in the world. Considering the high correlation between music, mood and driving comfort and safety, it makes sense to use appropriate and intelligent music recommendations based on the mood of drivers and songs in the context of car driving. The objective of this paper is to present the project of a contextual mood-based music recommender system capable of regulating the driver's mood and trying to have a positive influence on her driving behaviour. Here we present the proof of concept of the system and describe the techniques and technologies that are part of it. Further possible future improvements on each of the building blocks are also presented.

Erion Çano, Ondřej Bojar

Automatic methods and metrics that assess various quality criteria of automatically generated texts are important for developing NLG systems because they produce repeatable results and allow for a fast development cycle. We present here an attempt to automate the evaluation of text naturalness which is a very important characteristic of natural language generation methods. Instead of relying on human participants for scoring or labeling the text samples, we propose to automate the process by using a human likeliness metric we define and a discrimination procedure based on large pretrained language models with their probability distributions. We analyze the text probability fractions and observe how they are influenced by the size of the generative and discriminative models involved in the process. Based on our results, bigger generators and larger pretrained discriminators are more appropriate for a better evaluation of text naturalness. A comprehensive validation procedure with human participants is required as follow up to check how well this automatic evaluation scheme correlates with human judgments.

Erion Çano, Ondřej Bojar

Automatic evaluation of various text quality criteria produced by data-driven intelligent methods is very common and useful because it is cheap, fast, and usually yields repeatable results. In this paper, we present an attempt to automate the human likeliness evaluation of the output text samples coming from natural language generation methods used to solve several tasks. We propose to use a human likeliness score that shows the percentage of the output samples from a method that look as if they were written by a human. Instead of having human participants label or rate those samples, we completely automate the process by using a discrimination procedure based on large pretrained language models and their probability distributions. As follow up, we plan to perform an empirical analysis of human-written and machine-generated texts to find the optimal setup of this evaluation approach. A validation procedure involving human participants will also check how the automatic evaluation correlates with human judgments.

Erion Çano, Maurizio Morisio

Word embeddings or distributed representations of words are being used in various applications like machine translation, sentiment analysis, topic identification etc. Quality of word embeddings and performance of their applications depends on several factors like training method, corpus size and relevance etc. In this study we compare performance of a dozen of pretrained word embedding models on lyrics sentiment analysis and movie review polarity tasks. According to our results, Twitter Tweets is the best on lyrics sentiment analysis, whereas Google News and Common Crawl are the top performers on movie polarity analysis. Glove trained models slightly outrun those trained with Skipgram. Also, factors like topic relevance and size of corpus significantly impact the quality of the models. When medium or large-sized text sets are available, obtaining word embeddings from same training dataset is usually the best choice.

Erion Çano, Ondřej Bojar

Recent developments in sequence-to-sequence learning with neural networks have considerably improved the quality of automatically generated text summaries and document keywords, stipulating the need for even bigger training corpora. Metadata of research articles are usually easy to find online and can be used to perform research on various tasks. In this paper, we introduce two huge datasets for text summarization (OAGSX) and keyword generation (OAGKX) research, containing 34 million and 23 million records, respectively. The data were retrieved from the Open Academic Graph which is a network of research profiles and publications. We carefully processed each record and also tried several extractive and abstractive methods of both tasks to create performance baselines for other researchers. We further illustrate the performance of those methods previewing their outputs. In the near future, we would like to apply topic modeling on the two sets to derive subsets of research articles from more specific disciplines.

Erion Çano, Ondřej Bojar

Extractive keyphrase generation research has been around since the nineties, but the more advanced abstractive approach based on the encoder-decoder framework and sequence-to-sequence learning has been explored only recently. In fact, more than a dozen of abstractive methods have been proposed in the last three years, producing meaningful keyphrases and achieving state-of-the-art scores. In this survey, we examine various aspects of the extractive keyphrase generation methods and focus mostly on the more recent abstractive methods that are based on neural networks. We pay particular attention to the mechanisms that have driven the perfection of the later. A huge collection of scientific article metadata and the corresponding keyphrases is created and released for the research community. We also present various keyphrase generation and text summarization research patterns and trends of the last two decades.

Erion Çano, Ondřej Bojar

Using data-driven models for solving text summarization or similar tasks has become very common in the last years. Yet most of the studies report basic accuracy scores only, and nothing is known about the ability of the proposed models to improve when trained on more data. In this paper, we define and propose three data efficiency metrics: data score efficiency, data time deficiency and overall data efficiency. We also propose a simple scheme that uses those metrics and apply it for a more comprehensive evaluation of popular methods on text summarization and title generation tasks. For the latter task, we process and release a huge collection of 35 million abstract-title pairs from scientific articles. Our results reveal that among the tested models, the Transformer is the most efficient on both tasks.

Erion Çano, Ondřej Bojar

Authors' keyphrases assigned to scientific articles are essential for recognizing content and topic aspects. Most of the proposed supervised and unsupervised methods for keyphrase generation are unable to produce terms that are valuable but do not appear in the text. In this paper, we explore the possibility of considering the keyphrase string as an abstractive summary of the title and the abstract. First, we collect, process and release a large dataset of scientific paper metadata that contains 2.2 million records. Then we experiment with popular text summarization neural architectures. Despite using advanced deep learning models, large quantities of data and many days of computation, our systematic evaluation on four test datasets reveals that the explored text summarization methods could not produce better keyphrases than the simpler unsupervised methods, or the existing supervised ones.

Erion Çano, Maurizio Morisio

This work investigates the role of factors like training method, training corpus size and thematic relevance of texts in the performance of word embedding features on sentiment analysis of tweets, song lyrics, movie reviews and item reviews. We also explore specific training or post-processing methods that can be used to enhance the performance of word embeddings in certain tasks or domains. Our empirical observations indicate that models trained with multithematic texts that are large and rich in vocabulary are the best in answering syntactic and semantic word analogy questions. We further observe that influence of thematic relevance is stronger on movie and phone reviews, but weaker on tweets and lyrics. These two later domains are more sensitive to corpus size and training method, with Glove outperforming Word2vec. "Injecting" extra intelligence from lexicons or generating sentiment specific word embeddings are two prominent alternatives for increasing performance of word embedding features.

Erion Çano, Maurizio Morisio

Recommender systems are software tools used to generate and provide suggestions for items and other entities to the users by exploiting various strategies. Hybrid recommender systems combine two or more recommendation strategies in different ways to benefit from their complementary advantages. This systematic literature review presents the state of the art in hybrid recommender systems of the last decade. It is the first quantitative review work completely focused in hybrid recommenders. We address the most relevant problems considered and present the associated data mining and recommendation techniques used to overcome them. We also explore the hybridization classes each hybrid recommender belongs to, the application domains, the evaluation process and proposed future research directions. Based on our findings, most of the studies combine collaborative filtering with another technique often in a weighted way. Also cold-start and data sparsity are the two traditional and top problems being addressed in 23 and 22 studies each, while movies and movie datasets are still widely used by most of the authors. As most of the studies are evaluated by comparisons with similar methods using accuracy metrics, providing more credible and user oriented evaluations remains a typical challenge. Besides this, newer challenges were also identified such as responding to the variation of user context, evolving user tastes or providing cross-domain recommendations. Being a hot topic, hybrid recommenders represent a good basis with which to respond accordingly by exploring newer opportunities such as contextualizing recommendations, involving parallel hybrid algorithms, processing larger datasets, etc.

Erion Çano, Ondřej Bojar

In the area of online communication, commerce and transactions, analyzing sentiment polarity of texts written in various natural languages has become crucial. While there have been a lot of contributions in resources and studies for the English language, "smaller" languages like Czech have not received much attention. In this survey, we explore the effectiveness of many existing machine learning algorithms for sentiment analysis of Czech Facebook posts and product reviews. We report the sets of optimal parameter values for each algorithm and the scores in both datasets. We finally observe that support vector machines are the best classifier and efforts to increase performance even more with bagging, boosting or voting ensemble schemes fail to do so.

Erion Çano

Sentiment polarity of tweets, blog posts or product reviews has become highly attractive and is utilized in recommender systems, market predictions, business intelligence and more. Deep learning techniques are becoming top performers on analyzing such texts. There are however several problems that need to be solved for efficient use of deep neural networks on text mining and text polarity analysis. First, deep neural networks need to be fed with data sets that are big in size as well as properly labeled. Second, there are various uncertainties regarding the use of word embedding vectors: should they be generated from the same data set that is used to train the model or it is better to source them from big and popular collections? Third, to simplify model creation it is convenient to have generic neural network architectures that are effective and can adapt to various texts, encapsulating much of design complexity. This thesis addresses the above problems to provide methodological and practical insights for utilizing neural networks on sentiment analysis of texts and achieving state of the art results. Regarding the first problem, the effectiveness of various crowdsourcing alternatives is explored and two medium-sized and emotion-labeled song data sets are created utilizing social tags. To address the second problem, a series of experiments with large text collections of various contents and domains were conducted, trying word embeddings of various parameters. Regarding the third problem, a series of experiments involving convolution and max-pooling neural layers were conducted. Combining convolutions of words, bigrams, and trigrams with regional max-pooling layers in a couple of stacks produced the best results. The derived architecture achieves competitive performance on sentiment polarity analysis of movie, business and product reviews.