Now showing 1 - 8 of 8
  • Publication
    Machine Learning Techniques for Automatic Sensor Fault Detection in HUMS Systems
    (Engineers Australia, 2017-02-28) ; ;
    In this paper we describe the problem of developing sensor fault detection within HUMS instrumentation systems, and solutions based upon machine-learning techniques. We conclude with a report on our proof-of-concept demonstrator, and outline next-steps towards implementation of a autonomous self diagnostic sensor solution.
  • Publication
    Novel2Vec: Characterising 19th Century Fiction via Word Embeddings
    Recently, considerable attention has been paid to word embedding algorithms inspired by neural network models. Given a large textual corpus, these algorithms attempt to derive a set of vectors which represent the corpus vocabulary in a new embedded space. This representation can provide a useful means of measuring the underlying similarity between words. Here we investigate this property in the context of annotated texts of 19th-century fiction by the authors Jane Austen, Charles Dickens, and Arthur Conan Doyle. We demonstrate that building word embeddings on these texts can provide us with an insight into how characters group differently under different conditions, allowing us to make comparisons across different novels and authors. These results suggest that word embeddings can potentially provide a useful tool in supporting quantitative literary analysis.
  • Publication
    Windows on Waverley: exploring the effect of variations in the construction of literary social networks
    In recent years, social network analysis (SNA) has become increasingly popular as a quantitative approach to the examination of literary works, allowing researchers to generate abstract models of character groupings and interactions that appear in texts, and providing new opportunities for the evaluation of theories about communities and societies in literature. The social networks that are generated for a given novel, however, will differ considerably depending on what choices are made in relation to their construction: what types of interactions or co-occurrences are examined, what characters or other entities are considered, whether full texts or subsections such as chapters are investigated, and what automated methods are utilised for extracting character data, among others. This paper examines the effect of varying one specific aspect of network construction, by applying different "sliding window" strategies in order to create variations on social networks in three rather different early 19th-century novels: Pride and Prejudice (1813), Waverley (1814), and Frankenstein (1818). Three window strategies (collinear, co-planar and combination) are discussed, each of which captures qualitatively different social links between characters. We argue that the resulting networks yield different insights into a variety of aspects of the novels' construction, including narrative style and interactions between characters of different social class. We also suggest that rather than seeking to determine a single best-practice methodology for literary SNA, it may instead be illuminating to experiment with different approaches to the modelling of literary texts as social networks.
  • Publication
    Exploring the Role of Gender in 19th Century Fiction Through the Lens of Word Embeddings
    Within the last decade, substantial advances have been made in the field of computational linguistics, due in part to the evolution of word embedding algorithms inspired by neural network models. These algorithms attempt to derive a set of vectors which represent the vocabulary of a textual corpus in a new embedded space. This new representation can then be used to measure the underlying similarity between words. In this paper, we explore the role an author's gender may play in the selection of words that they choose to construct their narratives. Using a curated corpus of forty-eight 19th century novels, we generate, visualise, and investigate word embedding representations using a list of gender-encoded words. This allows us to explore the different ways in which male and female authors of this corpus use terms relating to contemporary understandings of gender and gender roles.
  • Publication
    Temporal Alignment of Reddit Network Embeddings
    Motivated by the concepts and findings being developed for diachronic word embeddings, in this paper, we explore how the application of the same principles can be leveraged to study structural roles from a temporal perspective. In the same way words with a similar meaning will repetitively appear in the same contexts, structural roles in graphs are also defined by the topological company that they keep. However, structurally equivalent roles may or may not occur in close proximity within a graph. Our goal is to map the participants of the popular social media website Reddit1, into an embedding space that best represents the similarity of the structural roles that they occupy and to then measure how their roles change over time.
  • Publication
    Machine Learning Techniques for Automatic Sensor Fault Detection in Airborne SHM Networks
    Good data is key to the success of a structural health monitoring (SHM) program, and modern data acquisition systems allow for reliable, high fidelity data capture. Unfortunately SHM programs are often hindered by undetected sensor and wiring problems resulting in invalid data. Many authors have identified sensors as the weakest link in an entire SHM system, where the transducer and the transducer/structure interface can ‘‘make or break’’ an SHM system. Choosing long-life sensors is one approach for addressing this problem, however these high cost and high specification sensors are rarely economically viable. For airborne monitoring programs, there is an expectation that sensors will be replaced over time, and that dedicated data analysts will be available to spot subtle signs within data which indicate the onset of sensor faults. This approach does not scale well for large fleets and does not allow for robust automation. In this paper, we present a new sensor diagnostic approach based on ‘‘Machine Learning’’ techniques. These automated techniques allow for reliable measurement using practical cost sensors, installed in extended duration monitoring programs. Machine Learning fault detection techniques not only detect the obvious catastrophic sensor errors, but also the more subtle sensor issues that can easily go undetected for long periods of time, e.g. strain gauge de-lamination, accelerometer de-calibration, loose/dry solder joints, adhesive degradation etc. A key enabler for airborne SHM systems is the automation of human data analysis, allowing systems to operate reliably without intervention for many years. This paper explores how Machine Learning techniques can be used to detect subtle signs of sensor/wiring faults within captured data, essentially automating the experience of human analysts who must ensure captured data is good. In today’s airborne data acquisition systems, built in diagnostics are typically limited in scope to a single subsystem and in functionality to a small number of failure scenarios. The Machine Learning techniques explored in this paper offer new expanded diagnostic capabilities which go beyond individual electronics units to provide system-wide diagnostics, encompassing all critical parts of an SHM system.
  • Publication
    Discovering Structure in Social Networks of 19th Century Fiction
    Inspired by the increasing availability of large text corpora online, digital humanities scholars are adopting computational approaches to explore questions in the field of literature from new perspectives. In this paper, we examine detailed social networks of characters, extracted from several works of 19th century fiction by Jane Austen and Charles Dickens. This allows us to apply methodologies from social network analysis, such as community detection, to explore the structure of these networks. By evaluating the results in collaboration with literary scholars, we find that the structure of the character networks can reveal underlying structural aspects within a novel, particularly in relation to plot and characterisation.
      729Scopus© Citations 10
  • Publication
    Temporal Analysis of Reddit Networks via Role Embeddings
    Inspired by diachronic word analysis from the field of natural language processing, we propose an approach for uncovering temporal insights regarding user roles from social networks using graph embedding methods. Specifically, we apply the role embedding algorithm, struc2vec, to a collection of social networks exhibiting either “loyal” or “vagrant” characteristics derived from the popular online social news aggregation website Reddit. For each subreddit, we extract nine months of data and create network role embeddings on consecutive time windows. We are then able to compare and contrast how user roles change over time by aligning the resulting temporal embeddings spaces. In particular, we analyse temporal role embeddings from an individual and a community-level perspective for both loyal and vagrant communities present on Reddit.