Template-Free Prompting for Few-Shot Named Entity Recognition via Semantic-Enhanced Contrastive Learning.

Prompt tuning has achieved great success in various sentence-level classification tasks by using elaborated label word mappings and prompt templates. However, for solving token-level classification tasks, e.g., named entity recognition (NER), previous research, which utilizes N-gram traversal for prompting all spans with all possible entity types, is time-consuming. To this end, we propose a novel prompt-based contrastive learning method for few-shot NER without template construction and label word mappings. First, we leverage external knowledge to initialize semantic anchors for each entity type. These anchors are simply appended with input sentence embeddings as template-free prompts (TFPs). Then, the prompts and sentence embeddings are in-context optimized with our proposed semantic-enhanced contrastive loss. Our proposed loss function enables contrastive learning in few-shot scenarios without requiring a significant number of negative samples. Moreover, it effectively addresses the issue of conventional contrastive learning, where negative instances with similar semantics are erroneously pushed apart in natural language processing (NLP)-related tasks. We examine our method in label extension (LE), domain-adaption (DA), and low-resource generalization evaluation tasks with six public datasets and different settings, achieving state-of-the-art (SOTA) results in most cases.

Survey on sentiment analysis: evolution of research methods and topics.

Sentiment analysis, one of the research hotspots in the natural language processing field, has attracted the attention of researchers, and research papers on the field are increasingly published. Many literature reviews on sentiment analysis involving techniques, methods, and applications have been produced using different survey methodologies and tools, but there has not been a survey dedicated to the evolution of research methods and topics of sentiment analysis. There have also been few survey works leveraging keyword co-occurrence on sentiment analysis. Therefore, this study presents a survey of sentiment analysis focusing on the evolution of research methods and topics. It incorporates keyword co-occurrence analysis with a community detection algorithm. This survey not only compares and analyzes the connections between research methods and topics over the past two decades but also uncovers the hotspots and trends over time, thus providing guidance for researchers. Furthermore, this paper presents broad practical insights into the methods and topics of sentiment analysis, while also identifying technical directions, limitations, and future work.

A Mixed Approach for Aggressive Political Discourse Analysis on Twitter.

Political tensions have grown throughout Europe since the beginning of the new century. The consecutive crises led to the rise of different social movements in several countries, in which the political status quo changed. These changes included an increment of the different tensions underlying politics, as has been reported after many other political and economical crises during the twentieth century. This article proposes the study of the political discourse, and its underlying tension, during Madrid's elections (Spain) in May 2021 by using a mixed approach. To demonstrate if an aggressive tone is used during the campaign, a mixed methodology approach is applied: quantitative computational techniques, related to natural language processing, are used to conduct a first general analysis of the information screened; then, these methods are used for detecting specific trends that can be later filtered and analyzed using a qualitative approach (content analysis), which is also conducted to extract insights about the information found. The main outcomes of this study show that the electoral campaign is not as negative as perceived by the citizens and that there was no relationship between the tone of the discourse and its dissemination. The analysis confirms that the most ideologically extreme parties tend to have a more aggressive language than the moderate ones. The content analysis carried out using our methodology showed that Twitter is used as a sentiment thermometer more than as a way of communicating concrete politics.

JCBIE: a joint continual learning neural network for biomedical information extraction.

Extracting knowledge from heterogeneous data sources is fundamental for the construction of structured biomedical knowledge graphs (BKGs), where entities and relations are represented as nodes and edges in the graphs, respectively. Previous biomedical knowledge extraction methods simply considered limited entity types and relations by using a task-specific training set, which is insufficient for large-scale BKGs development and downstream task applications in different scenarios. To alleviate this issue, we propose a joint continual learning biomedical information extraction (JCBIE) network to extract entities and relations from different biomedical information datasets. By empirically studying different joint learning and continual learning strategies, the proposed JCBIE can learn and expand different types of entities and relations from different datasets. JCBIE uses two separated encoders in joint-feature extraction, hence can effectively avoid the feature confusion problem comparing with using one hard-parameter sharing encoder. Specifically, it allows us to adopt entity augmented inputs to establish the interaction between named entity recognition and relation extraction. Finally, a novel evaluation mechanism is proposed for measuring cross-corpus generalization errors, which was ignored by traditional evaluation methods. Our empirical studies show that JCBIE achieves promising performance when continual learning strategy is adopted with multiple corpora.

Disentangled Retrieval and Reasoning for Implicit Question Answering.

To date, most of the existing open-domain question answering (QA) methods focus on explicit questions where the reasoning steps are mentioned explicitly in the question. In this article, we study implicit QA where the reasoning steps are not evident in the question. Implicit QA is challenging in two aspects. First, evidence retrieval is difficult since there is little overlap between a question and its required evidence. Second, answer inference is difficult since the reasoning strategy is latent in the question. To tackle implicit QA, we propose a systematic solution denoted as DisentangledQA, which disentangles topic, attribute, and reasoning strategy from the implicit question to guide the retrieval and reasoning. Specifically, we disentangle the topic and attribute information from the implicit question to guide evidence retrieval. For answer reasoning, we propose a disentangled reasoning model for answer prediction based on retrieved evidence as well as the latent representation of the reasoning strategy. The disentangled framework empowers each module to focus on a specific latent element in the question, and thus, leads to effective representation learning for them. Experiments on the StrategyQA dataset demonstrate the effectiveness of our method in answering implicit questions, improving performance in evidence retrieval and answering inference by 31.7% and 4.5%, respectively, and achieving the best performance on the official leaderboard. In addition, our method achieved the best performance on the challenging EntityQuestions dataset, indicating the effectiveness in improving general open-domain QA tasks.

Mood of the Planet: Challenging Visions of Big Data in the Arts.

Mood of the Planet is an interactive physical-digital sculpture that has as its center-piece a large "arch" or "doorway" that emits colored light and sound as a form of visualization and sonification of the changing, live emotions expressed by people all around the Earth. It is the product of several disciplines, including the arts, computer science, linguistics and psychology. In particular, we use artificial intelligence to collect and analyze social media data and extract emotions from these using a brain-inspired and psychologically motivated emotion categorization model. Such emotions are then translated into colors and sounds that the audience can experience while passing through the arch. Feedback from the audience proved the Mood of the Planet to provide a more accurate, personal and tangible experience about the data-emotions dichotomy.

A Survey on Knowledge Graphs: Representation, Acquisition, and Applications.

Human knowledge provides a formal understanding of the world. Knowledge graphs that represent structural relations between entities have become an increasingly popular research direction toward cognition and human-level intelligence. In this survey, we provide a comprehensive review of the knowledge graph covering overall research topics about: 1) knowledge graph representation learning; 2) knowledge acquisition and completion; 3) temporal knowledge graph; and 4) knowledge-aware applications and summarize recent breakthroughs and perspective directions to facilitate future research. We propose a full-view categorization and new taxonomies on these topics. Knowledge graph embedding is organized from four aspects of representation space, scoring function, encoding models, and auxiliary information. For knowledge acquisition, especially knowledge graph completion, embedding methods, path inference, and logical rule reasoning are reviewed. We further explore several emerging topics, including metarelational learning, commonsense reasoning, and temporal knowledge graphs. To facilitate future research on knowledge graphs, we also provide a curated collection of data sets and open-source libraries on different tasks. In the end, we have a thorough outlook on several promising research directions.

Emotion Recognition on Edge Devices: Training and Deployment.

Emotion recognition, among other natural language processing tasks, has greatly benefited from the use of large transformer models. Deploying these models on resource-constrained devices, however, is a major challenge due to their computational cost. In this paper, we show that the combination of large transformers, as high-quality feature extractors, and simple hardware-friendly classifiers based on linear separators can achieve competitive performance while allowing real-time inference and fast training. Various solutions including batch and Online Sequential Learning are analyzed. Additionally, our experiments show that latency and performance can be further improved via dimensionality reduction and pre-training, respectively. The resulting system is implemented on two types of edge device, namely an edge accelerator and two smartphones.

Graph routing between capsules.

Routing methods in capsule networks often learn a hierarchical relationship for capsules in successive layers, but the intra-relation between capsules in the same layer is less studied, while this intra-relation is a key factor for the semantic understanding in text data. Therefore, in this paper, we introduce a new capsule network with graph routing to learn both relationships, where capsules in each layer are treated as the nodes of a graph. We investigate strategies to yield adjacency and degree matrix with three different distances from a layer of capsules, and propose the graph routing mechanism between those capsules. We validate our approach on five text classification datasets, and our findings suggest that the approach combining bottom-up routing and top-down attention performs the best. Such an approach demonstrates generalization capability across datasets. Compared to the state-of-the-art routing methods, the improvements in accuracy in the five datasets we used were 0.82, 0.39, 0.07, 1.01, and 0.02, respectively.

Conversational Memory Network for Emotion Recognition in Dyadic Dialogue Videos.

Emotion recognition in conversations is crucial for the development of empathetic machines. Present methods mostly ignore the role of inter-speaker dependency relations while classifying emotions in conversations. In this paper, we address recognizing utterance-level emotions in dyadic conversational videos. We propose a deep neural framework, termed conversational memory network, which leverages contextual information from the conversation history. The framework takes a multimodal approach comprising audio, visual and textual features with gated recurrent units to model past utterances of each speaker into memories. Such memories are then merged using attention-based hops to capture inter-speaker dependencies. Experiments show an accuracy improvement of 3-4% over the state of the art.

Multi-attention Recurrent Network for Human Communication Comprehension.

Human face-to-face communication is a complex multimodal signal. We use words (language modality), gestures (vision modality) and changes in tone (acoustic modality) to convey our intentions. Humans easily process and understand face-to-face communication, however, comprehending this form of communication remains a significant challenge for Artificial Intelligence (AI). AI must understand each modality and the interactions between them that shape the communication. In this paper, we present a novel neural architecture for understanding human communication called the Multi-attention Recurrent Network (MARN). The main strength of our model comes from discovering interactions between modalities through time using a neural component called the Multi-attention Block (MAB) and storing them in the hybrid memory of a recurrent component called the Long-short Term Hybrid Memory (LSTHM). We perform extensive comparisons on six publicly available datasets for multimodal sentiment analysis, speaker trait recognition and emotion recognition. MARN shows state-of-the-art results performance in all the datasets.

Multilingual Sentiment Analysis: State of the Art and Independent Comparison of Techniques.

With the advent of Internet, people actively express their opinions about products, services, events, political parties, etc., in social media, blogs, and website comments. The amount of research work on sentiment analysis is growing explosively. However, the majority of research efforts are devoted to English-language data, while a great share of information is available in other languages. We present a state-of-the-art review on multilingual sentiment analysis. More importantly, we compare our own implementation of existing approaches on common data. Precision observed in our experiments is typically lower than the one reported by the original authors, which we attribute to the lack of detail in the original presentation of those approaches. Thus, we compare the existing works by what they really offer to the reader, including whether they allow for accurate implementation and for reliable reproduction of the reported results.

Role of Muscle Synergies in Real-Time Classification of Upper Limb Motions using Extreme Learning Machines.

BACKGROUND: Myoelectric signals offer significant insights in interpreting the motion intention and extent of effort involved in performing a movement, with application in prostheses, orthosis and exoskeletons. Feature extraction plays a vital role, and follows two approaches: EMG and synergy features. More recently, muscle synergy based features are being increasingly explored, since it simplifies dimensionality of control, and are considered to be more robust to signal variations. Another important aspect in a myoelectrically controlled devices is the learning capability and speed of performance for online decoding. Extreme learning machine (ELM) is a relatively new neural-network based learning algorithm: its performance hasn't been explored in the context of online control, which is a more reliable measure compared to offline analysis. To this purpose we aim at focusing our investigation on a myoelectric-based interface which is able to identify and online classify, upper limb motions involving shoulder and elbow. The main objective is to compare the performance of the decoder trained using ELM, for two different features: EMG and synergy features. METHODS: The experiments are broadly divided in two phases training/calibration and testing respectively. ELM is used to train the decoder using data acquired during the calibration phase. The performance of the decoder is then tested in online motion control by using a simulated graphical user interface replicating the human limb: subjects are requested to control a virtual arm by using their muscular activity. The decoder performance is quantified using ad-hoc metrics based on the following indicators: motion selection time, motion completion time, and classification accuracy. RESULTS: Performance has been evaluated for both offline and online contexts. The offline classification results indicated better performance in the case of EMG features, whereas a better classification accuracy for synergy feature was observed for online decoding. Also the other indicators as motion selection time and motion completion time, showed better trend in the case of synergy than time-domain features. CONCLUSION: This work demonstrates better robustness of online decoding of upper-limb motions and motor intentions when using synergy feature. Furthermore, we have quantified the performance of the decoder trained using ELM for online control, providing a potential and viable option for real-time myoelectric control in assistive technology.

Muscle synergies for reliable classification of arm motions using myoelectric interface.

Synergistic activation of muscles are considered to be the phenomenon by which the central nervous system simplifies its control strategy. Muscle synergies are neurally encoded and considered robust to be able to adapt for various external dynamics. This paper presents a myoelectric-based interface to identify and classify motions of the upper arm involving the shoulder and elbow. We contrast performance of the decoder while using time domain and synergy features. The decoder is trained using extreme learning machine algorithm, and online testing is performed in a virtual environment. Better classification accuracy for online control is obtained while using muscle synergy features. The results indicate better online performance compared to offline performance while using synergy features to classify movements, indicating generalization to dynamic situations and robustness of control.

Towards an intelligent framework for multimodal affective data analysis.

An increasingly large amount of multimodal content is posted on social media websites such as YouTube and Facebook everyday. In order to cope with the growth of such so much multimodal data, there is an urgent need to develop an intelligent multi-modal analysis framework that can effectively extract information from multiple modalities. In this paper, we propose a novel multimodal information extraction agent, which infers and aggregates the semantic and affective information associated with user-generated multimodal data in contexts such as e-learning, e-health, automatic video content tagging and human-computer interaction. In particular, the developed intelligent agent adopts an ensemble feature extraction approach by exploiting the joint use of tri-modal (text, audio and video) features to enhance the multimodal information extraction process. In preliminary experiments using the eNTERFACE dataset, our proposed multi-modal system is shown to achieve an accuracy of 87.95%, outperforming the best state-of-the-art system by more than 10%, or in relative terms, a 56% reduction in error rate.

Enhancing business intelligence by means of suggestive reviews.

Appropriate identification and classification of online reviews to satisfy the needs of current and potential users pose a critical challenge for the business environment. This paper focuses on a specific kind of reviews: the suggestive type. Suggestions have a significant influence on both consumers' choices and designers' understanding and, hence, they are key for tasks such as brand positioning and social media marketing. The proposed approach consists of three main steps: (1) classify comparative and suggestive sentences; (2) categorize suggestive sentences into different types, either explicit or implicit locutions; (3) perform sentiment analysis on the classified reviews. A range of supervised machine learning approaches and feature sets are evaluated to tackle the problem of suggestive opinion mining. Experimental results for all three tasks are obtained on a dataset of mobile phone reviews and demonstrate that extending a bag-of-words representation with suggestive and comparative patterns is ideal for distinguishing suggestive sentences. In particular, it is observed that classifying suggestive sentences into implicit and explicit locutions works best when using a mixed sequential rule feature representation. Sentiment analysis achieves maximum performance when employing additional preprocessing in the form of negation handling and target masking, combined with sentiment lexicons.