Digital Commons Network^™

A Tidy Data Model For Natural Language Processing Using Cleannlp, Taylor Arnold

The R Journal

Recent advances in natural language processing have produced libraries that extract low level features from a collection of raw texts. These features, known as annotations, are usually stored internally in hierarchical, tree-based data structures. This paper proposes a data model to represent annotations as a collection of normalized relational data tables optimized for exploratory data analysis and predictive modeling. The R package cleanNLP, which calls one of two state of the art NLP libraries (CoreNLP or spaCy), is presented as an implementation of this data model. It takes raw text as an input and returns a list of normalized …

Building An Albanian Knowledge Graph From Unstructured Text, Hakan Shehu

Theses and Dissertations

We live in the golden area of information. The World Wide web contains a vast amount of unstructured text in different digital formats, including newswire, blogs, email communications, governmental documents, chat logs, and so on. Some of the biggest companies and organizations have created knowledge bases which represent a semantic network of facts, entities and relations between them. Even though this area has been well researched for a considerable time, there is a lack of implementation of such a knowledge extraction for Albanian language. In this thesis we will try to create an Albanian general knowledge graph from unstructured text. …

Predicting The Author Of Twitter Posts With Markov Chain Analysis, Daniel Freeman

Honors Theses

Given a set of text with known authors, is it possible to take new text, not knowing who wrote it, and correctly identify the author? One way to do this is to analyze the text using Markov chains. This research project will first attempt to answer this question using books available in the public domain. Using what is learned from trying to identify authors of books, the primary goal of this project is to identify the best way to guess the author of a post on the social media network Twitter using Markov chains.

A Tidy Data Model For Natural Language Processing Using Cleannlp, Taylor B. Arnold

Department of Math & Statistics Faculty Publications

Recent advances in natural language processing have produced libraries that extract low level features from a collection of raw texts. These features, known as annotations, are usually stored internally in hierarchical, tree-based data structures. This paper proposes a data model to represent annotations as a collection of normalized relational data tables optimized for exploratory data analysis and predictive modeling. The R package cleanNLP, which calls one of two state of the art NLP libraries (CoreNLP or spaCy), is presented as an implementation of this data model. It takes raw text as an input and returns a list of normalized tables. …

Leveraging Auxiliary Tasks For Document-Level Cross-Domain Sentiment Classification, Jianfei Yu, Jing Jiang

Research Collection School Of Computing and Information Systems

In this paper, we study domain adaptationwith a state-of-the-art hierarchicalneural network for document-level sentimentclassification. We first design a newauxiliary task based on sentiment scoresof domain-independent words. We thenpropose two neural network architecturesto respectively induce document embeddingsand sentence embeddings that workwell for different domains. When thesedocument and sentence embeddings areused for sentiment classification, we findthat with both pseudo and external sentimentlexicons, our proposed methods canperform similarly to or better than severalhighly competitive domain adaptationmethods on a benchmark dataset of productreviews.

Leveraging The Trade-Off Between Accuracy And Interpretability In A Hybrid Intelligent System, Di Wang, Chai Quek, Ah-Hwee Tan, Chunyan Miao, Geok See Ng, You Zhou

Research Collection School Of Computing and Information Systems

Neural Fuzzy Inference System (NFIS) is a widely adopted paradigm to develop a data-driven learning system. This hybrid system has been widely adopted due to its accurate reasoning procedure and comprehensible inference rules. Although most NFISs primarily focus on accuracy, we have observed an ever increasing demand on improving the interpretability of NFISs and other types of machine learning systems. In this paper, we illustrate how we leverage the trade-off between accuracy and interpretability in an NFIS called Genetic Algorithm and Rough Set Incorporated Neural Fuzzy Inference System (GARSINFIS). In a nutshell, GARSINFIS self-organizes its network structure with a small …

Authorship Identification Of Translation Algorithms., Keishin Nishiyama

Electronic Theses and Dissertations

Authorship analysis is a process of identifying a true writer of a given document and has been studied for decades. However, only a handful of studies of authorship analysis of translators are available despite the fact that online translations are widely available and also popularly employed in automatic translations of posts in social networking services. The identification of translation algorithms has potential to contribute to the investigation of cybercrimes, involving translation of scam messages by algorithmic translations to reach speakers of foreign languages. This study tested bag of words (BOW) approach in authorship attribution and the existing approaches to translator …

Inferring Social Media Users’ Demographics From Profile Pictures: A Face++ Analysis On Twitter Users, Soon-Gyo Jung, Jisun An, Haewoon Kwak, Joni Salminen, Bernard J. Jansen

Research Collection School Of Computing and Information Systems

In this research, we evaluate the applicability of using facial recognition of social media account profile pictures to infer the demographic attributes of gender, race, and age of the account owners leveraging a commercial and well-known image service, specifically Face++. Our goal is to determine the feasibility of this approach for actual system implementation. Using a dataset of approximately 10,000 Twitter profile pictures, we use Face++ to classify this set of images for gender, race, and age. We determine that about 30% of these profile pictures contain identifiable images of people using the current state-of-the-art automated means. We then employ …

A Framework For Clustering And Adaptive Topic Tracking On Evolving Text And Social Media Data Streams., Gopi Chand Nutakki

Electronic Theses and Dissertations

Recent advances and widespread usage of online web services and social media platforms, coupled with ubiquitous low cost devices, mobile technologies, and increasing capacity of lower cost storage, has led to a proliferation of Big data, ranging from, news, e-commerce clickstreams, and online business transactions to continuous event logs and social media expressions. These large amounts of online data, often referred to as data streams, because they get generated at extremely high throughputs or velocity, can make conventional and classical data analytics methodologies obsolete. For these reasons, the issues of management and analysis of data streams have been researched extensively …

The Growth, Structure, And Dynamics Of Crime In Cities, Marcos Oliveira

Theses and Dissertations

Crime is ubiquitous in cities but lacks a statistical characterization that could lead to uncovering its underlying mechanisms. Cities are, however, in a constant process of organization making difficult to analyze urban phenomena. Yet, to understand urbanization and its consequences, we need to approach cities as evolving processes, instead of static objects. With this perspective, we examined regularities in crime regarding its growth, structure, and dynamics. We developed frameworks to examine the spatial, temporal, and periodic variations of crime in cities. Though thefts increase super-linearly with the population, we found burglaries showing a linear increase. Our analyses also confirmed crime …

On Modeling Sense Relatedness In Multi-Prototype Word Embedding, Yixin Cao, Juanzi Li, Jiaxin Shi, Zhiyuan Liu, Chengjiang Li

Research Collection School Of Computing and Information Systems

To enhance the expression ability of distributional word representation learning model, many researchers tend to induce word senses through clustering, and learn multiple embedding vectors for each word, namely multi-prototype word embedding model. However, most related work ignores the relatedness among word senses which actually plays an important role. In this paper, we propose a novel approach to capture word sense relatedness in multi-prototype word embedding model. Particularly, we differentiate the original sense and extended senses of a word by introducing their global occurrence information and model their relatedness through the local textual context information. Based on the idea of …

Extracting Implicit Suggestions From Students’ Comments: A Text Analytics Approach, Venky Shankararaman, Swapna Gottipati, Jeff Rongsheng Lin, Sandy Gan

Research Collection School Of Computing and Information Systems

At the end of each course, students are required to give feedback on the course and instructor. This feedback includes quantitative rating using Likert scale and qualitative feedback as comments. Such qualitative feedback can provide valuable insights in helping the instructor enhance the course content and teaching delivery. However, the main challenge in analysing the qualitative feedback is the perceived increase in time and effort needed to manually process the textual comments. In this paper, we provide an automated solution for analysing comments, specifically extracting implicit suggestions from the students’ qualitative feedback comments. The implemented solution leverages existing text mining …

Multilingual Sentiment Analysis : From Formal To Informal And Scarce Resource Languages, Siaw Ling Lo, Erik Cambria, Raymond Chiong, David Cornforth

Research Collection School Of Computing and Information Systems

The ability to analyse online user-generated content related to sentiments (e.g., thoughts and opinions) on products or policies has become a de-facto skillset for many companies and organisations. Besides the challenge of understanding formal textual content, it is also necessary to take into consideration the informal and mixed linguistic nature of online social media languages, which are often coupled with localised slang as a way to express ‘true’ feelings. Due to the multilingual nature of social media data, analysis based on a single official language may carry the risk of not capturing the overall sentiment of online content. While efforts …

Modular Mechanistic Networks: On Bridging Mechanistic And Phenomenological Models With Deep Neural Networks In Natural Language Processing, Simon Dobnik, John D. Kelleher

Books/Book chapters

Natural language processing (NLP) can be done using either top-down (theory driven) and bottom-up (data driven) approaches, which we call mechanistic and phenomenological respectively. The approaches are frequently considered to stand in opposition to each other. Examining some recent approaches in deep learning we argue that deep neural networks incorporate both perspectives and, furthermore, that leveraging this aspect of deep learning may help in solving complex problems within language technology, such as modelling language and perception in the domain of spatial cognition.

What Is Not Where: The Challenge Of Integrating Spatial Representations Into Deep Learning Architectures, John D. Kelleher, Simon Dobnik

Books/Book chapters

This paper examines to what degree current deep learning architectures for image caption generation capture spatial lan- guage. On the basis of the evaluation of examples of generated captions from the literature we argue that systems capture what objects are in the image data but not where these objects are located: the cap- tions generated by these systems are the output of a language model conditioned on the output of an object detector that cannot capture fine-grained location information. Although language models provide useful knowledge for image captions, we argue that deep learning image captioning architectures should also model geometric …

Perfect Ratings With Negative Comments: Learning From Contradictory Patient Survey Responses, Andrew S. Gallan, Marina Girju, Roxana Girju

Patient Experience Journal

This research explores why patients give perfect domain scores yet provide negative comments on surveys. In order to explore this phenomenon, vendor-supplied in-patient survey data from eleven different hospitals of a major U.S. health care system were utilized. The dataset included survey scores and comments from 56,900 patients, collected from January 2015 through October 2016. Of the total number of responses, 30,485 (54%) contained at least one comment. For our analysis, we use a two-step approach: a quantitative analysis on the domain scores augmented by a qualitative text analysis of patients’ comments. To focus the research, we start by building …

Ordinal Convolutional Neural Networks For Predicting Rdoc Positive Valence Psychiatric Symptom Severity Scores, Anthony Rios, Ramakanth Kavuluru

Computer Science Faculty Publications

Background—The CEGS N-GRID 2016 Shared Task in Clinical Natural Language Processing (NLP) provided a set of 1000 neuropsychiatric notes to participants as part of a competition to predict psychiatric symptom severity scores. This paper summarizes our methods, results, and experiences based on our participation in the second track of the shared task.

Objective—Classical methods of text classification usually fall into one of three problem types: binary, multi-class, and multi-label classification. In this effort, we study ordinal regression problems with text data where misclassifications are penalized differently based on how far apart the ground truth and model predictions are …

Detecting Semantic Uncertainty By Learning Hedge Cues In Sentences Using An Hmm, Xiujun Li, Wei Gao, Jude Shavlik

Research Collection School Of Computing and Information Systems

Detecting speculative assertions is essential to distinguish semantically uncertain information from the factual ones in text. This is critical to the trustworthiness of many intelligent systems that are based on information retrieval and natural language processing techniques, such as question answering or information extraction. We empirically explore three fundamental issues of uncertainty detection: (1) the predictive ability of different learning methods on this task; (2) whether using unlabeled data can lead to a more accurate model; and (3) whether closed-domain training or crossdomain training is better. For these purposes, we adopt two statistical learning approaches to this problem: the commonly …

Sourcevote: Fusing Multi-Valued Data Via Inter-Source Agreements, Xiu Susie Fang, Quan Z. Sheng, Xianzhi Wang, Mahmoud Barhamgi, Lina Yao, Anne H.H. Ngu

Research Collection School Of Computing and Information Systems

Data fusion is a fundamental research problem of identifying true values of data items of interest from conflicting multi-sourced data. Although considerable research efforts have been conducted on this topic, existing approaches generally assume every data item has exactly one true value, which fails to reflect the real world where data items with multiple true values widely exist. In this paper, we propose a novel approach,SourceVote, to estimate value veracity for multi-valued data items. SourceVote models the endorsement relations among sources by quantifying their two-sided inter-source agreements. In particular, two graphs are constructed to model inter-source relations. Then two aspects …

Sourcevote: Fusing Multi-Valued Data Via Inter-Source Agreements, Xiu Susie Fang, Quan Z. Sheng, Xianzhi Wang, Mahmoud Barhamgi, Lina Yao, Anne H.H. Ngu

Research Collection School Of Computing and Information Systems

Data fusion is a fundamental research problem of identifyingtrue values of data items of interest from conflicting multi-sourceddata. Although considerable research efforts have been conducted on thistopic, existing approaches generally assume every data item has exactlyone true value, which fails to reflect the real world where data items withmultiple true values widely exist. In this paper, we propose a novel approach,SourceVote, to estimate value veracity for multi-valued data items.SourceVote models the endorsement relations among sources by quantifyingtheir two-sided inter-source agreements. In particular, two graphs areconstructed to model inter-source relations. Then two aspects of sourcereliability are derived from these graphs and …

Improving Probability Estimation Through Active Probabilistic Model Learning, Jingyi Wang, Xiaohong Chen, Jun Sun, Shengchao Qin

Research Collection School Of Computing and Information Systems

It is often necessary to estimate the probability of certain events occurring in a system. For instance, knowing the probability of events triggering a shutdown sequence allows us to estimate the availability of the system. One approach is to run the system multiple times and then construct a probabilistic model to estimate the probability. When the probability of the event to be estimated is low, many system runs are necessary in order to generate an accurate estimation. For complex cyber-physical systems, each system run is costly and time-consuming, and thus it is important to reduce the number of system runs …

Single Versus Concurrent Systems: Nominal Classification In Mian, Greville G. Corbett, Sebastian Fedden, Raphael Finkel

Computer Science Faculty Publications

The Papuan language Mian allows us to refine the typology of nominal classification. Mian has two candidate classification systems, differing completely in their formal realization but overlapping considerably in their semantics. To determine whether to analyse Mian as a single system or concurrent systems we adopt a canonical approach. Our criteria – orthogonality of the systems (we give a precise measure), semantic compositionality, morphosyntactic alignment, distribution across parts of speech, exponence, and interaction with other features – point mainly to an analysis as concurrent systems. We thus improve our analysis of Mian and make progress with the typology of nominal …

Critical Digital Literacies Across Scales And Beneath The Screen, Noah Asher Golden

Education Faculty Articles and Research

Digital technologies and education scholarship tend to focus on either individual creative design or analysis of the political economy. To better understand how ideologies travel across networks, critical digital literacies must focus on enactments beneath the screen, as the linguistic constructs known as software can enact interests across scales of activity to “disembed” local actions and meaning. Investigations of these mobilities and disembedding effects challenge popular notions of digital technologies as neutral, rendering overt the ways that algorithms can naturalize manifestations of power and social arrangements. Such a framework allows for descriptive analyses of the ways hegemonic discourses are enacted …

Cloud‐Based Text Analytics Harvesting, Cleaning And Analyzing Corporate Earnings Conference Calls, Michael Chuancai Zhang, Vikram Gazula, Dan Stone, Hong Xie

Commonwealth Computational Summit

No abstract provided.

Adversarial Discriminative Domain Adaptation For Extracting Protein-Protein Interactions From Text, Anthony Rios, Ramakanth Kavuluru, Zhiyong Lu

Commonwealth Computational Summit

Relation extraction is the process of extracting structured information from unstructured text. Recently, neural networks (NNs) have produced state-of-art results in extracting protein-protein interactions (PPIs) from text. While multiple corpora have been created to extract PPIs from text, most methods have shown poor cross-corpora generalization. In other words, models trained on one dataset perform poorly on other datasets for the same task. In the case of PPI, the F1 has been shown to vary by as much as 30% between different datasets. In this work, we utilize adversarial discriminative domain adaptation (ADDA) to improve the generalization between the source and …

Cloud-Based Text Analytics: Harvesting, Cleaning And Analyzing Corporate Earnings Conference Calls, Michael Chuancai Zhang, Vikram Gazula, Dan Stone, Hong Xie

Commonwealth Computational Summit

Does management language cohesion in earnings conference calls matter to the capital market? As a part of the research on the above question, and taking advantage of the modern IT technologies, this project:

• harvested 115,882 earnings conference call transcripts from SeekingAlpha.com
• parsed and structured 89,988 transcripts using regular expressions in Stata
• analyzed 179,976 text files using Amazon Elastic Compute Cloud (Amazon EC2), which
• saved almost 2 years (675 days) of the project time

What Does Explainable Ai Really Mean? A New Conceptualization Of Perspectives, Derek Doran, Sarah Schulz, Tarek R. Besold

Computer Science and Engineering Faculty Publications

We characterize three notions of explainable AI that cut across research fields: opaque systems that offer no insight into its algo- rithmic mechanisms; interpretable systems where users can mathemat- ically analyze its algorithmic mechanisms; and comprehensible systems that emit symbols enabling user-driven explanations of how a conclusion is reached. The paper is motivated by a corpus analysis of NIPS, ACL, COGSCI, and ICCV/ECCV paper titles showing differences in how work on explainable AI is positioned in various fields. We close by introducing a fourth notion: truly explainable systems, where automated reasoning is central to output crafted explanations without requiring human …

Word Sense Determination From Wikipedia Data Using Neural Networks, Qiao Liu

Master's Projects

Many words have multiple meanings. For example, “plant” can mean a type of living organism or a factory. Being able to determine the sense of such words is very useful in natural language processing tasks, such as speech synthesis, question answering, and machine translation. For the project described in this report, we used a modular model to classify the sense of words to be disambiguated. This model consisted of two parts: The first part was a neural-network-based language model to compute continuous vector representations of words from data sets created from Wikipedia pages. The second part classified the meaning of …

Improve And Implement An Open Source Question Answering System, Salil Shenoy

Master's Projects

A question answer system takes queries from the user in natural language and returns a short concise answer which best fits the response to the question. This report discusses the integration and implementation of question answer systems for English and Hindi as part of the open source search engine Yioop. We have implemented a question answer system for English and Hindi, keeping in mind users who use these languages as their primary language. The user should be able to query a set of documents and should get the answers in the same language. English and Hindi are very different when …

A Conceptual Framework For Analyzing Students' Feedback, Venky Shankararaman, Swapna Gottipati, Sandy Gan

Research Collection School Of Computing and Information Systems

In academic institutions it is normal practice that at the end of each term,students are required to complete a questionnaire that is designed to gather students’perceptions of the instructor and their learning experience in the course. This questionnaire comprises of Likert-scale questions and qualitative questions.One of the important goals of this exercise is to enable the instructor and the senior management to examine the feedback and then enhance students’ learning experience. In most universities, including our own, a lot of attention is paid to the quantitative feedback, which is summarized and statistical comparisons are computed, analysed and presented. However, the …