This study exploits the power of collective inference and proposes an HFGM based on heterogeneous entity networks, which significantly improved our ability to extract relations from TCM literature.

BackgroundWe aim to automatically extract species names of bacteria and their locations from webpages. This task is important for exploiting the vast amount of biological knowledge which is expressed in diverse natural language texts and putting this knowledge in databases for easy access by biologists. The task is challenging and the previous results are far below an acceptable level of performance, particularly for extraction of localization relationships. Therefore, we aim to design a new system for such extractions, using the framework of structured machine learning techniques.ResultsWe design a new model for joint extraction of biomedical entities and the localization relationship. Our model is based on a spatial role labeling (SpRL) model designed for spatial understanding of unrestricted text. We extend SpRL to extract discourse level spatial relations in the biomedical domain and apply it on the BioNLP-ST 2013, BB-shared task. We highlight the main differences between general spatial language understanding and spatial information extraction from the scientific text which is the focus of this work. We exploit the text’s structure and discourse level global features. Our model and the designed features substantially improve on the previous systems, achieving an absolute improvement of approximately 57 percent over F1 measure of the best previous system for this task.ConclusionsOur experimental results indicate that a joint learning model over all entities and relationships in a document outperforms a model which extracts entities and relationships independently. Our global learning model significantly improves the state-of-the-art results on this task and has a high potential to be adopted in other natural language processing (NLP) tasks in the biomedical domain.

Abstract. Traditional summarization initiatives have been focused on specific types of documents such as articles, reviews, videos, image feeds, or tweets, a practice which may result in pigeonholing the summarization task in the context of modern, content-rich multimedia collections. Consequently, much of the research to date has revolved around mostly toy problems in narrow domains and working on single-source media types. We argue that summarization and story generation systems need to refocus the problem space in order to meet the information needs in the age of user-generated content in di↵erent formats and languages. Here we create a framework for flexible multimedia storytelling. Narratives, stories, and summaries carry a set of challenges in big data and dynamic multi-source media that give rise to new research in spatial-temporal representation, viewpoint generation, and explanation.

When we refer to an image that attracts our attention, it is natural to mention not only what is literally depicted in the image, but also the sentiments, thoughts and opinions that it invokes in ourselves. In this work we deviate from the standard mainstream tasks of associating tags or keywords to an image, or generating content image descriptions, and we introduce the novel task of automatically generate user comments for an image. We present a new dataset collected from the social media Pinterest and we propose a strategy based on building joint textual and visual user models, tailored to the specificity of the mentioned task. We conduct an extensive experimental analysis of our approach on both qualitative and quantitative terms, which allows assessing the value of the proposed approach and shows its encouraging results against several existing image-to-text methods.

We explore the benefits that multitask learning offer to speech processing as we train models on dual objectives with automatic speech recognition and intent classification or sentiment classification. Our models, although being of modest size, show improvements over models trained end-to-end on intent classification. We compare different settings to find the optimal disposition of each task module compared to one another. Finally, we study the performance of the models in lowresource scenario by training the models with as few as one example per class. We show that multitask learning in these scenarios compete with a baseline model trained on text features and performs considerably better than a pipeline model. On sentiment classification, we match the performance of an end-to-end model with ten times as many parameters. We consider 4 tasks and 4 datasets in Dutch and English.