This paper describes the submission of team EURECOM at SemEval-2024 Task 4: Multilingual Detection of Persuasion Techniques in Memes. We only tackled the first sub-task, consisting of detecting 20 named persuasion techniques in the textual content of memes. We trained multiple BERT-based models (BERT, RoBERTa, BERT pre-trained on harmful detection) using different losses (Cross Entropy, Binary Cross Entropy, Focal Loss and a custom-made hierarchical loss). The best results were obtained by leveraging the hierarchical nature of the data, by outputting ancestor classes and with a hierarchical loss. Our final submission consist of an ensembling of our top-3 best models for each persuasion techniques. We obtain hierarchical F1 scores of 0.655 (English), 0.345 (Bulgarian), 0.442 (North Macedonian) and 0.178 (Arabic) on the test set.

This paper presents D2KLab’s system used for the shared task of “Multilingual Complex Named Entity Recognition (MultiCoNER II)”, as part of SemEval 2023 Task 2. The system relies on a fine-tuned transformer based language model for extracting named entities. In addition to the architecture of the system, we discuss our results and observations.

Large language models have recently risen in popularity due to their ability to perform many natural language tasks without requiring any fine-tuning. In this work, we focus on two novel ideas: (1) generating definitions from examples and using them for zero-shot classification, and (2) investigating how an LLM makes use of the definitions. We thoroughly analyze the performance of GPT-3 model for fine-grained multi-label conspiracy theory classification of tweets using zero-shot labeling. In doing so, we asses how to improve the labeling by providing minimal but meaningful context in the form of the definitions of the labels. We compare descriptive noun phrases, human-crafted definitions, introduce a new method to help the model generate definitions from examples, and propose a method to evaluate GPT-3’s understanding of the definitions. We demonstrate that improving definitions of class labels has a direct consequence on the downstream classification results.

We present a benchmark in six European languages containing manually annotated information about olfactory situations and events following a FrameNet-like approach. The documents selection covers ten domains of interest to cultural historians in the olfactory domain and includes texts published between 1620 to 1920, allowing a diachronic analysis of smell descriptions. With this work, we aim to foster the development of olfactory information extraction approaches as well as the analysis of changes in smell descriptions over time.

From statistical to neural models, a wide variety of topic modelling algorithms have been proposed in the literature. However, because of the diversity of datasets and metrics, there have not been many efforts to systematically compare their performance on the same benchmarks and under the same conditions. In this paper, we present a selection of 9 topic modelling techniques from the state of the art reflecting a diversity of approaches to the task, an overview of the different metrics used to compare their performance, and the challenges of conducting such a comparison. We empirically evaluate the performance of these models on different settings reflecting a variety of real-life conditions in terms of dataset size, number of topics, and distribution of topics, following identical preprocessing and evaluation processes. Using both metrics that rely on the intrinsic characteristics of the dataset (different coherence metrics), as well as external knowledge (word embeddings and ground-truth topic labels), our experiments reveal several shortcomings regarding the common practices in topic models evaluation.

The knowledge of the European silk textile production is a typical case for which the information collected is heterogeneous, spread across many museums and sparse since rarely complete. Knowledge Graphs for this cultural heritage domain, when being developed with appropriate ontologies and vocabularies, enable to integrate and reconcile this diverse information. However, many of these original museum records still have some metadata gaps. In this paper, we present a zero-shot learning approach that leverages the ConceptNet common sense knowledge graph to predict categorical metadata informing about the silk objects production. We compared the performance of our approach with traditional supervised deep learning-based methods that do require training data. We demonstrate promising and competitive performance for similar datasets and circumstances and the ability to predict sometimes more fine-grained information. Our results can be reproduced using the code and datasets published at https://github.com/silknow/ZSL-KG-silk.

Computational fact-checking has gained a lot of traction in the machine learning and natural language processing communities. A plethora of solutions have been developed, but methods which leverage both structured and unstructured information to detect misinformation are of particular relevance. In this paper, we tackle the FEVEROUS (Fact Extraction and VERification Over Unstructured and Structured information) challenge which consists of an open source baseline system together with a benchmark dataset containing 87,026 verified claims. We extend this baseline model by improving the evidence retrieval module yielding the best evidence F1 score among the competitors in the challenge leaderboard while obtaining an overall FEVEROUS score of 0.20 (5th best ranked system).

From LDA to neural models, different topic modeling approaches have been proposed in the literature. However, their suitability and performance is not easy to compare, particularly when the algorithms are being used in the wild on heterogeneous datasets. In this paper, we introduce ToModAPI (TOpic MOdeling API), a wrapper library to easily train, evaluate and infer using different topic modeling algorithms through a unified interface. The library is extensible and can be used in Python environments or through a Web API.

Entity linking systems typically rely on encyclopedic knowledge bases such as DBpedia or Freebase. In this paper, we use, instead, a French lexical-semantic network named JeuxDeMots to jointly type and link entities. Our approach combines word embeddings and a path-based similarity resulting in encouraging results over a set of documents from the French Le Monde newspaper.

This paper describes the MeMAD project entry to the WMT Multimodal Machine Translation Shared Task. We propose adapting the Transformer neural machine translation (NMT) architecture to a multi-modal setting. In this paper, we also describe the preliminary experiments with text-only translation systems leading us up to this choice. We have the top scoring system for both English-to-German and English-to-French, according to the automatic metrics for flickr18. Our experiments show that the effect of the visual features in our system is small. Our largest gains come from the quality of the underlying text-only NMT system. We find that appropriate use of additional data is effective.

In this paper, we describe the participation of the SentiME++ system to the SemEval 2017 Task 4A “Sentiment Analysis in Twitter” that aims to classify whether English tweets are of positive, neutral or negative sentiment. SentiME++ is an ensemble approach to sentiment analysis that leverages stacked generalization to automatically combine the predictions of five state-of-the-art sentiment classifiers. SentiME++ achieved officially 61.30% F1-score, ranking 12th out of 38 participants.

More and more knowledge bases are publicly available as linked data. Since these knowledge bases contain structured descriptions of real-world entities, they can be exploited by entity linking systems that anchor entity mentions from text to the most relevant resources describing those entities. In this paper, we investigate adaptation of the entity linking task using contextual knowledge. The key intuition is that entity linking can be customized depending on the textual content, as well as on the application that would make use of the extracted information. We present an adaptive approach that relies on contextual knowledge from text to enhance the performance of ADEL, a hybrid linguistic and graph-based entity linking system. We evaluate our approach on a domain-specific corpus consisting of annotated WikiNews articles.

Named entity recognition and disambiguation are of primary importance for extracting information and for populating knowledge bases. Detecting and classifying named entities has traditionally been taken on by the natural language processing community, whilst linking of entities to external resources, such as those in DBpedia, has been tackled by the Semantic Web community. As these tasks are treated in different communities, there is as yet no oversight on the performance of these tasks combined. We present an approach that combines the state-of-the art from named entity recognition in the natural language processing domain and named entity linking from the semantic web community. We report on experiments and results to gain more insights into the strengths and limitations of current approaches on these tasks. Our approach relies on the numerous web extractors supported by the NERD framework, which we combine with a machine learning algorithm to optimize recognition and linking of named entities. We test our approach on four standard data sets that are composed of two diverse text types, namely newswire and microposts.