Isabelle Augenstein

LODifier: generating linked data from unstructured text

The automated extraction of information from text and its transformation into a formal description is an important goal in both Semantic Web research and computational linguistics. The extracted information can be used for a variety of tasks such as ontology generation, question answering and information retrieval. LODifier is an approach that combines deep semantic analysis with named entity recognition, word sense disambiguation and controlled Semantic Web vocabularies in order to extract named entities and relations between them from text and to convert them into an RDF representation which is linked to DBpedia and WordNet. We present the architecture of our tool and discuss design decisions made. An evaluation of the tool on a story link detection task gives clear evidence of its practical potential.

Unsupervised wrapper induction using linked data

This work explores the usage of Linked Data for Web scale Information Extraction and shows encouraging results on the task of Wrapper Induction. We propose a simple knowledge based method which is (i) highly flexible with respect to different domains and (ii) does not require any training material, but exploits Linked Data as background knowledge source to build essential learning resources. The major contribution of this work is a study of how Linked Data - an imprecise, redundant and large-scale knowledge resource - can be used to support Web scale Information Extraction in an effective and efficient way and identify the challenges involved. We show that, for domains that are covered, Linked Data serve as a powerful knowledge resource for Information Extraction. Experiments on a publicly available dataset demonstrate that, under certain conditions, this simple unsupervised approach can achieve competitive results against some complex state of the art that always depends on training data.

emoji2vec: Learning Emoji Representations from their Description.

Many current natural language processing applications for social media rely on representation learning and utilize pre-trained word embeddings. There currently exist several publicly-available, pre-trained sets of word embeddings, but they contain few or no emoji representations even as emoji usage in social media has increased. In this paper we release emoji2vec, pre-trained embeddings for all Unicode emoji which are learned from their description in the Unicode emoji standard. The resulting emoji embeddings can be readily used in downstream social natural language processing applications alongside word2vec. We demonstrate, for the downstream task of sentiment analysis, that emoji embeddings learned from short descriptions outperforms a skip-gram model trained on a large collection of tweets, while avoiding the need for contexts in which emoji need to appear frequently in order to estimate a representation.

Relation Extraction from the Web Using Distant Supervision

Extracting information from Web pages requires the ability to work at Web scale in terms of the number of documents, the number of domains and domain complexity. Recent approaches have used existing knowledge bases to learn to extract information with promising results. In this paper we propose the use of distant supervision for relation extraction from the Web. Distant supervision is a method which uses background information from the Linking Open Data cloud to automatically label sentences with relations to create training data for relation classifiers. Although the method is promising, existing approaches are still not suitable for Web extraction as they suffer from three main issues: data sparsity, noise and lexical ambiguity. Our approach reduces the impact of data sparsity by making entity recognition tools more robust across domains, as well as extracting relations across sentence boundaries. We reduce the noise caused by lexical ambiguity by employing statistical methods to strategically select training data. Our experiments show that using a more robust entity recognition approach and expanding the scope of relation extraction results in about 8 times the number of extractions, and that strategically selecting training data can result in an error reduction of about 30%.

Statistical Knowledge Patterns: Identifying Synonymous Relations in Large Linked Datasets

The Web of Data is a rich common resource with billions of triples available in thousands of datasets and individual Web documents created by both expert and non-expert ontologists. A common problem is the imprecision in the use of vocabularies: annotators can misunderstand the semantics of a class or property or may not be able to find the right objects to annotate with. This decreases the quality of data and may eventually hamper its usability over large scale. This paper describes Statistical Knowledge Patterns (SKP) as a means to address this issue. SKPs encapsulate key information about ontology classes, including synonymous properties in (and across) datasets, and are automatically generated based on statistical data analysis. SKPs can be effectively used to automatically normalise data, and hence increase recall in querying. Both pattern extraction and pattern usage are completely automated. The main benefits of SKPs are that: (1) their structure allows for both accurate query expansion and restriction; (2) they are context dependent, hence they describe the usage and meaning of properties in the context of a particular class; and (3) they can be generated offline, hence the equivalence among relations can be used efficiently at run time.

USFD at SemEval-2016 Task 6: Any-Target Stance Detection on Twitter with Autoencoders

This paper describes the University of Sheffield’s submission to the SemEval 2016 Twitter Stance Detection weakly supervised task (SemEval 2016 Task 6, Subtask B). In stance detection, the goal is to classify the stance of a tweet towards a target as “favor”, “against”, or “none”. In Subtask B, the targets in the test data are different from the targets in the training data, thus rendering the task more challenging but also more realistic. To address the lack of target-specific training data, we use a large set of unlabelled tweets containing all targets and train a bag-of-words autoencoder to learn how to produce feature representations of tweets. These feature representations are then used to train a logistic regression classifier on labelled tweets, with additional features such as an indicator of whether the target is contained in the tweet. Our submitted run on the test data achieved an F1 of 0.3270.

Distantly supervised Web relation extraction for knowledge base population

Extracting information from Web pages for populating large, cross-domain knowledge bases requires methods which are suitable across domains, do not require manual effort to adapt to new domains, are able to deal with noise, and integrate information extracted from different Web pages. Recent approaches have used existing knowledge bases to learn to extract information with promising results, one of those approaches being distant supervision. Distant supervision is an unsupervised method which uses background information from the Linking Open Data cloud to automatically label sentences with relations to create training data for relation classifiers. In this paper we propose the use of distant supervision for relation extraction from the Web. Although the method is promising, existing approaches are still not suitable for Web extraction as they suffer from three main issues: data sparsity, noise and lexical ambiguity. Our approach reduces the impact of data sparsity by making entity recognition tools more robust across domains and extracting relations across sentence boundaries using unsupervised co- reference resolution methods. We reduce the noise caused by lexical ambiguity by employing statistical methods to strategically select training data. To combine information extracted from multiple sources for populating knowledge bases we present and evaluate several information integration strategies and show that those benefit immensely from additional relation mentions extracted using co-reference resolution, increasing precision by 8%. We further show that strategically selecting training data can increase precision by a further 3%.

Mapping Keywords to Linked Data Resources for Automatic Query Expansion

Linked Data is a gigantic, constantly growing and extremely valuable resource, but its usage is still heavily dependent on (i) the familiarity of end users with RDF’s graph data model and its query language, SPARQL, and (ii) knowledge about available datasets and their contents. Intelligent keyword search over Linked Data is currently being investigated as a means to overcome these barriers to entry in a number of different approaches, including semantic search engines and the automatic conversion of natural language questions into structured queries. Our work addresses the specific challenge of mapping keywords to Linked Data resources, and proposes a novel method for this task. By exploiting the graph structure within Linked Data we determine which properties between resources are useful to discover, or directly express, semantic similarity. We also propose a novel scoring function to rank results. Experiments on a publicly available dataset show a 17% improvement in Mean Reciprocal Rank over the state of the art.

Extracting Relations between Non-Standard Entities using Distant Supervision and Imitation Learning

Distantly supervised approaches have become popular in recent years as they allow training relation extractors without textbound annotation, using instead known relations from a knowledge base and a large textual corpus from an appropriate domain. While state of the art distant supervision approaches use off-theshelf named entity recognition and classification (NERC) systems to identify relation arguments, discrepancies in domain or genre between the data used for NERC training and the intended domain for the relation extractor can lead to low performance. This is particularly problematic for “non-standard” named entities such as album which would fall into the MISC category. We propose to ameliorate this issue by jointly training the named entity classifier and the relation extractor using imitation learning which reduces structured prediction learning to classification learning. We further experiment with Web features different features and compare against using two off-the-shelf supervised NERC systems, Stanford NER and FIGER, for named entity classification. Our experiments show that imitation learning improves average precision by 4 points over an one-stage classification model, while removing Web features results in a 6 points reduction. Compared to using FIGER and Stanford NER, average precision is 10 points and 19 points higher with our imitation learning approach.

USFD: Twitter NER with Drift Compensation and Linked Data

This paper describes a pilot NER system for Twitter, comprising the USFD system entry to the W-NUT 2015 NER shared task. The goal is to correctly label entities in a tweet dataset, using an inventory of ten types. We employ structured learning, drawing on gazetteers taken from Linked Data, and on unsupervised clustering features, and attempting to compensate for stylistic and topic drift - a key challenge in social media text. Our result is competitive; we provide an analysis of the components of our methodology, and an examination of the target dataset in the context of this task.