David Milne is a lecturer in the School of Information, Systems and Modelling. His interests are in Information Retrieval and Text Mining, and in designing, developing and evaluating web and mobile applications that put these algorithms into the hands of end-users. He is most widely known for his work in mining Wikipedia, but has more recently focused on the domain of mental health.
He believes that services such as online counselling and online peer support could greatly benefit from AI to help ensure they remain safe, timely and theraputic at scale. However, he also recognizes that there are many challenges and potential pitfalls in applying automation in this highly sensitive context. Conseqently his research aims to ensure AI systems complement and facilitate interactions between people rather than replacing them.
David's work has received numerous awards, including a 2018 Test of Time Award and the 2008 Best Interdisciplinary Paper Award from the ACM International Conference on Information and Knowledge Management, the 2015 Most Cited Award from the International Journal of Human-Computer Studies, and an honourable mention for the 2011 CORE Australasian Distinguished Doctoral Dissertation Award.
Can supervise: YES
Milne, DN, McCabe, KL & Calvo, RA 2019, 'Improving moderator responsiveness in online peer support through automated triage', Journal of Medical Internet Research, vol. 21, no. 4.View/Download from: UTS OPUS or Publisher's site
© 2019 Journal of Medical Internet Research. All rights reserved. Background: Online peer support forums require oversight to ensure they remain safe and therapeutic. As online communities grow, they place a greater burden on their human moderators, which increases the likelihood that people at risk may be overlooked. This study evaluated the potential for machine learning to assist online peer support by directing moderators' attention where it is most needed. Objective: This study aimed to evaluate the accuracy of an automated triage system and the extent to which it influences moderator behavior. Methods: A machine learning classifier was trained to prioritize forum messages as green, amber, red, or crisis depending on how urgently they require attention from a moderator. This was then launched as a set of widgets injected into a popular online peer support forum hosted by ReachOut.com, an Australian Web-based youth mental health service that aims to intervene early in the onset of mental health problems in young people. The accuracy of the system was evaluated using a holdout test set of manually prioritized messages. The impact on moderator behavior was measured as response ratio and response latency, that is, the proportion of messages that receive at least one reply from a moderator and how long it took for these replies to be made. These measures were compared across 3 periods: before launch, after an informal launch, and after a formal launch accompanied by training. Results: The algorithm achieved 84% f-measure in identifying content that required a moderator response. Between prelaunch and post-training periods, response ratios increased by 0.9, 4.4, and 10.5 percentage points for messages labelled as crisis, red, and green, respectively, but decreased by 5.0 percentage points for amber messages. Logistic regression indicated that the triage system was a significant contributor to response ratios for green, amber, and red messages, but not for crisis m...
Naqshbandi, K, Hoermann, S, Milne, D, Peters, D, Davies, B, Potter, S & Calvo, RA 2019, 'Codesigning technology for a voluntarysector organization', Human Technology, vol. 15, no. 1, pp. 6-29.View/Download from: UTS OPUS or Publisher's site
© 2019 Khushnood Naqshbandi, Simon Hoermann, David Milne, Dorian Peters, Benjamin Davies, Sophie Potter, & Rafael A. Calvo. This paper presents an investigation into the experiences and perceptions of volunteers and community managers of an Australian voluntary-sector organization that supports young help-seeking people. The process focused specifically on the design of a chat tool, a rudimentary version of which was conceptualized and tested during a trial completed prior to this study. The process explored the motivations and experiences of these volunteers using a codesign approach, which led to the development of specific features of the chat tool that were tailored to the nature of their work and organization, as well as the sector-specific ethos. We employed several research methods, which included interviews, focus groups, and participatory design workshops. Thematic analyses were performed on the resultant qualitative data. The methods, motivational themes, and the ensuing design solutions that were implemented are discussed in detail with the aim of encouraging codesign of technology for voluntary-sector organizations.
Choi, I, Milne, DN, Deady, M, Calvo, RA, Harvey, SB & Glozier, N 2018, 'Impact of Mental Health Screening on Promoting Immediate Online Help-Seeking: Randomized Trial Comparing Normative Versus Humor-Driven Feedback.', JMIR Mental Health, vol. 5, no. 2, pp. 1-7.View/Download from: UTS OPUS or Publisher's site
BACKGROUND:Given the widespread availability of mental health screening apps, providing personalized feedback may encourage people at high risk to seek help to manage their symptoms. While apps typically provide personal score feedback only, feedback types that are user-friendly and increase personal relevance may encourage further help-seeking. OBJECTIVE:The aim of this study was to compare the effects of providing normative and humor-driven feedback on immediate online help-seeking, defined as clicking on a link to an external resource, and to explore demographic predictors that encourage help-seeking. METHODS:An online sample of 549 adults were recruited using social media advertisements. Participants downloaded a smartphone app known as "Mindgauge" which allowed them to screen their mental wellbeing by completing standardized measures on Symptoms (Kessler 6-item Scale), Wellbeing (World Health Organization [Five] Wellbeing Index), and Resilience (Brief Resilience Scale). Participants were randomized to receive normative feedback that compared their scores to a reference group or humor-driven feedback that presented their scores in a relaxed manner. Those who scored in the moderate or poor ranges in any measure were encouraged to seek help by clicking on a link to an external online resource. RESULTS:A total of 318 participants scored poorly on one or more measures and were provided with an external link after being randomized to receive normative or humor-driven feedback. There was no significant difference of feedback type on clicking on the external link across all measures. A larger proportion of participants from the Wellbeing measure (170/274, 62.0%) clicked on the links than the Resilience (47/179, 26.3%) or Symptoms (26/75, 34.7%) measures (χ2=60.35, P<.001). There were no significant demographic factors associated with help-seeking for the Resilience or Wellbeing measures. Participants with a previous episode of poor mental health were less likely tha...
Deady, M, Johnston, DA, Glozier, N, Milne, DN, Choi, I, Mackinnon, A, Mykletun, A, Calvo, RA, Gayed, A, Bryant, R, Christensen, H & Harvey, SB 2018, 'A smartphone application for treating depressive symptoms: study protocol for a randomised controlled trial', BMC Psychiatry, vol. 18, no. 166, pp. 1-9.View/Download from: UTS OPUS or Publisher's site
Depression is a commonly occurring disorder linked to diminished role functioning and quality of life. The development of treatments that overcome barriers to accessing treatment remains an important area of clinical research as most people delay or do not receive treatment at an appropriate time. The workplace is an ideal setting to roll-out an intervention, particularly given the substantial psychological benefits associated with remaining in the workforce. Mobile health (mhealth) interventions utilising smartphone applications (apps) offer novel solutions to disseminating evidence based programs, however few apps have undergone rigorous testing. The present study aims to evaluate the effectiveness of a smartphone app designed to treat depressive symptoms in workers.
Deady, M, Johnston, DA, Glozier, N, Milne, D, Choi, I, Mackinnon, A, Mykletun, A, Calvo, RA, Gayed, A, Bryant, R, Christensen, H & Harvey, SB 2018, 'Smartphone application for preventing depression: Study protocol for a workplace randomised controlled trial', BMJ Open, vol. 8, no. 7, pp. 1-11.View/Download from: UTS OPUS or Publisher's site
© Author(s) (or their employer(s)) 2018. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ. Introduction Depression is the leading cause of life years lost due to disability. Appropriate prevention has the potential to reduce the incidence of new cases of depression, however, traditional prevention approaches face significant scalability issues. Prevention programmes delivered by via smartphone applications provide a potential solution. The workplace is an ideal setting to roll out this form of intervention, particularly among industries that are unlikely to access traditional health initiatives and whose workplace characteristics create accessibility and portability issues. The study aims to evaluate the effectiveness of a smartphone application designed to prevent depression and improve well-being. The effectiveness of the app as a universal, selective and indicated prevention tool will also be evaluated. Methods and analysis A multicentre randomised controlled trial, to determine the effectiveness of the intervention compared with an active mood monitoring control in reducing depressive symptoms (primary outcome) and the prevalence of depression at 3 months, with secondary outcomes assessing well-being and work performance. Employees from a range of industries will be invited to participate. Participants with likely current depression at baseline will be excluded. Following baseline assessment, participants, blinded to their allocation, will be randomised to receive one of two versions of the application: headgear (a 30-day mental health intervention) or a control application (mood monitoring for 30 days). Both versions of the app contain a risk calculator to provide a measure of future risk. Analyses will be conducted within an intention-To-Treat framework using mixed modelling, with additional analyses conducted to compare the moderating effect of baseline risk level and depression symptom severity on the int...
Deady, M, Johnston, D, Milne, D, Glozier, N, Peters, D, Calvo, R & Harvey, S 2018, 'Preliminary Effectiveness of a Smartphone App to Reduce Depressive Symptoms in the Workplace: Feasibility and Acceptability Study.', JMIR mHealth and uHealth, vol. 6, no. 12, pp. 1-14.View/Download from: UTS OPUS or Publisher's site
BACKGROUND:The workplace represents a unique setting for mental health interventions. Due to range of job-related factors, employees in male-dominated industries are at an elevated risk. However, these at-risk groups are often overlooked. HeadGear is a smartphone app-based intervention designed to reduce depressive symptoms and increase well-being in these populations. OBJECTIVE:This paper presents the development and pilot testing of the app's usability, acceptability, feasibility, and preliminary effectiveness. METHODS:The development process took place from January 2016 to August 2017. Participants for prototype testing (n=21; stage 1) were recruited from industry partner organizations to assess acceptability and utility. A 5-week effectiveness and feasibility pilot study (n=84; stage 2) was then undertaken, utilizing social media recruitment. Demographic data, acceptability and utility questionnaires, depression (Patient Health Questionnaire-9), and other mental health measures were collected. RESULTS:The majority of respondents felt HeadGear was easy to use (92%), easily understood (92%), were satisfied with the app (67%), and would recommend it to a friend (75%; stage 1). Stage 2 found that compared with baseline, depression and anxiety symptoms were significantly lower at follow-up (t30=2.53; P=.02 and t30=2.18; P=.04, respectively), days of sick leave in past month (t28=2.38; P=.02), and higher self-reported job performance (t28=-2.09; P=.046; stage 2). Over 90% of respondents claimed it helped improve their mental fitness, and user feedback was again positive. Attrition was high across the stages. CONCLUSIONS:Overall, HeadGear was well received, and preliminary findings indicate it may provide an innovative new platform for improving mental health outcomes. Unfortunately, attrition was a significant issue, and findings should be interpreted with caution. The next stage of evaluation will be a randomized controlled trial. If found to be efficacious, the a...
Choi, I, Milne, DN, Glozier, N, Peters, D, Harvey, SB & Calvo, RA 2017, 'Using different Facebook advertisements to recruit men for an online mental health study: Engagement and selection bias', Internet Interventions, vol. 8, pp. 27-34.View/Download from: UTS OPUS or Publisher's site
© 2017 A growing number of researchers are using Facebook to recruit for a range of online health, medical, and psychosocial studies. There is limited research on the representativeness of participants recruited from Facebook, and the content is rarely mentioned in the methods, despite some suggestion that the advertisement content affects recruitment success. This study explores the impact of different Facebook advertisement content for the same study on recruitment rate, engagement, and participant characteristics. Five Facebook advertisement sets ('resilience', 'happiness', 'strength', 'mental fitness', and 'mental health') were used to recruit male participants to an online mental health study which allowed them to find out about their mental health and wellbeing through completing six measures. The Facebook advertisements recruited 372 men to the study over a one month period. The cost per participant from the advertisement sets ranged from $0.55 to $3.85 Australian dollars. The 'strength' advertisements resulted in the highest recruitment rate, but participants from this group were least engaged in the study website. The 'strength' and 'happiness' advertisements recruited more younger men. Participants recruited from the 'mental health' advertisements had worse outcomes on the clinical measures of distress, wellbeing, strength, and stress. This study confirmed that different Facebook advertisement content leads to different recruitment rates and engagement with a study. Different advertisement also leads to selection bias in terms of demographic and mental health characteristics. Researchers should carefully consider the content of social media advertisements to be in accordance with their target population and consider reporting this to enable better assessment of generalisability.
Calvo, RA, Milne, DN, Hussain, MS & Christensen, H 2017, 'Natural language processing in mental health applications using non-clinical texts', Natural Language Engineering, vol. 23, no. 5, pp. 649-685.View/Download from: UTS OPUS or Publisher's site
© Copyright Cambridge University Press 2017. Natural language processing (NLP) techniques can be used to make inferences about peoples' mental states from what they write on Facebook, Twitter and other social media. These inferences can then be used to create online pathways to direct people to health information and assistance and also to generate personalized interventions. Regrettably, the computational methods used to collect, process and utilize online writing data, as well as the evaluations of these techniques, are still dispersed in the literature. This paper provides a taxonomy of data sources and techniques that have been used for mental health support and intervention. Specifically, we review how social media and other data sources have been used to detect emotions and identify people who may be in need of psychological assistance; the computational techniques used in labeling and diagnosis; and finally, we discuss ways to generate and personalize mental health interventions. The overarching aim of this scoping review is to highlight areas of research where NLP has been applied in the mental health literature and to help develop a common language that draws together the fields of mental health, human-computer interaction and NLP.
Hoermann, S, McCabe, KL, Milne, DN & Calvo, RA 2017, 'Application of synchronous text-based dialogue systems in mental health interventions: Systematic review', Journal of Medical Internet Research, vol. 19, no. 8.View/Download from: UTS OPUS or Publisher's site
© Simon Hoermann, Kathryn L McCabe, David N Milne, Rafael A Calvo. Background: Synchronous written conversations (or "chats") are becoming increasingly popular as Web-based mental health interventions. Therefore, it is of utmost importance to evaluate and summarize the quality of these interventions. Objective: The aim of this study was to review the current evidence for the feasibility and effectiveness of online one-on-one mental health interventions that use text-based synchronous chat. Methods: A systematic search was conducted of the databases relevant to this area of research (Medical Literature Analysis and Retrieval System Online [MEDLINE], PsycINFO, Central, Scopus, EMBASE, Web of Science, IEEE, and ACM). There were no specific selection criteria relating to the participant group. Studies were included if they reported interventions with individual text-based synchronous conversations (ie, chat or text messaging) and a psychological outcome measure. Results: A total of 24 articles were included in this review. Interventions included a wide range of mental health targets (eg, anxiety, distress, depression, eating disorders, and addiction) and intervention design. Overall, compared with the waitlist (WL) condition, studies showed significant and sustained improvements in mental health outcomes following synchronous text-based intervention, and post treatment improvement equivalent but not superior to treatment as usual (TAU) (eg, face-to-face and telephone counseling). Conclusions: Feasibility studies indicate substantial innovation in this area of mental health intervention with studies utilizing trained volunteers and chatbot technologies to deliver interventions. While studies of efficacy show positive post-intervention gains, further research is needed to determine whether time requirements for this mode of intervention are feasible in clinical practice.
Huang, L, Milne, D, Frank, E & Witten, IH 2012, 'Learning a concept-based document similarity measure', JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY, vol. 63, no. 8, pp. 1593-1608.View/Download from: Publisher's site
Calvo, RA, Hussain, MS, Milne, D, Nordbo, K, Hickie, I & Danckwerts, P 2016, 'Augmenting online mental health support services' in Gaming and Technology Addiction: Breakthroughs in Research and Practice, pp. 264-285.View/Download from: Publisher's site
© 2017 by IGI Global. All rights reserved. The Internet allows mental health organisations to provide services to more people via new models of care. Amongst these, online support groups are becoming increasingly popular. This model of mental health service provision includes moderators who read posts, recognise those that need attention and provide support via online responses. However, as these groups become more popular they risk becoming more difficult to manage due to the sheer volume of posts. This challenge can be addressed through computational linguistics techniques. This chapter reports on work with a mental health organisation on three components to help scale up the number of people they can support. The design aims to go beyond helping end-users and explores how design can support the wellbeing of the moderators themselves. The design of the three components is discussed: 1) A triage component automatically detects posts that need a prompt response. 2) An intervention generator (IG) generates a draft response for the moderator to use, for example a positive psychology intervention. These two can help in the management of a discussion forum, supporting positive behaviours, not just dealing in situations of distress. 3) A component for synchronous support through an augmented chat system.
Altszyler, E, Berenstein, AJ, Milne, D, Calvo, RA & Fernandez Slezak, D 1970, 'Using contextual information for automatic triage of posts in a peer-support forum', Proceedings of the Fifth Workshop on Computational Linguistics and Clinical Psychology: From Keyboard to Clinic, Proceedings of the Fifth Workshop on Computational Linguistics and Clinical Psychology: From Keyboard to Clinic, Association for Computational Linguistics.View/Download from: UTS OPUS or Publisher's site
Naqshbandi, K, Milne, DN, Davies, B, Potter, S, Calvo, RA & Hoermann, S 2016, 'Helping young people going through tough times: Perspectives for a peer-to-peer chat support system', PROCEEDINGS OF THE 28TH AUSTRALIAN COMPUTER-HUMAN INTERACTION CONFERENCE (OZCHI 2016), 28th Australian Computer-Human Interaction Conference (OzCHI), ASSOC COMPUTING MACHINERY, Univ Tasmania, Hobart, AUSTRALIA.View/Download from: Publisher's site
Ijaz, K, Wang, Y, Milne, D & Calvo, RA 2016, 'Competitive vs Affiliative Design of Immersive VR Exergames', SERIOUS GAMES, JCSG 2016, 2nd International Joint Conference on Serious Games (JCSG), SPRINGER INTERNATIONAL PUBLISHING AG, Griffith Univ, Brisbane, AUSTRALIA, pp. 140-150.View/Download from: Publisher's site
Ijaz, K, Wang, Y, Milne, D & Calvo, RA 2016, 'VR-Rides: Interactive VR Games for Health', SERIOUS GAMES, JCSG 2016, 2nd International Joint Conference on Serious Games (JCSG), SPRINGER INTERNATIONAL PUBLISHING AG, Griffith Univ, Brisbane, AUSTRALIA, pp. 289-292.View/Download from: Publisher's site
Milne, DN, Pink, G, Hachey, B & Calvo, RA 1970, 'CLPsych 2016 Shared Task: Triaging content in online peer-support forums', Proceedings of the Third Workshop on Computational Lingusitics and Clinical Psychology, Proceedings of the Third Workshop on Computational Lingusitics and Clinical Psychology, Association for Computational Linguistics.View/Download from: UTS OPUS or Publisher's site
Nordbo, K, Milne, D, Calvo, RA & Allman-Farinelli, M 2015, 'Virtual food court: A VR environment to assess people's food choices', OzCHI 2015: Being Human - Conference Proceedings, pp. 69-72.View/Download from: Publisher's site
Copyright is held by the owner/author(s). Immersive virtual reality environments can provide users with realistic experiences of worlds that do not exist or would be hard to reach. The ability to manipulate these environments and influence experiences can be used to understand decision making under different conditions. In this study we explore how VR can be used to understand more about people's food choices. We explore how policy-based interventions such as the "sugar tax" and "nutrition labelling" to promote healthier food choices could be tested. Only limited experimental studies have been conducted about such choices due to the difficulty of trying such interventions in large retail settings. The objectives of the study were to assess how accurately the Virtual Food Court (VFC), represents a real food court. The study (27 participants) had two study conditions; a control with regular food-court prices, and an experimental condition with taxes on food and beverages. Results revealed that participants were able to imagine doing their real-life food purchases in the VFC indicating that it is a good research tool for assessing people's food choices.
This paper introduces HMpara, a new search engine that aims to make Wikipedia easier to explore. It works on top of the encyclopedia's existing link structure, abstracting away from document content and allowing users to navigate the resource at a higher level. It utilizes semantic relatedness measures to emphasize articles and connections that are most likely to be of interest, visualization to expose the structure of how the available information is organized, and lightweight information extraction to explain itself. © 2011 ACM.
Huang, A, Milne, D, Frank, E & Witten, IH 2009, 'Clustering Documents Using a Wikipedia-Based Concept Representation', ADVANCES IN KNOWLEDGE DISCOVERY AND DATA MINING, PROCEEDINGS, 13th Pacific-Asia Conference on Knowledge Discovery and Data Mining, SPRINGER-VERLAG BERLIN, Bangkok, THAILAND, pp. 628-636.
Medelyan, O & Milne, D 2008, 'Augmenting domain-specific thesau with knowledge from wikipedia', New Zealand Computer Science Research Student Conference, NZCSRSC 2008 - Proceedings, pp. 108-114.
We propose a new method for extending a domain-specific thesaurus with valuable information from Wikipedia. The main obstacle is to disambiguate thesaurus concepts to correct Wikipedia articles. Given the concept name, we first identify candidate mappings by analyzing article titles, their redirects and disambiguation pages. Then, for each candidate, we compute a link-based similarity score to all mappings of context terms related to this concept. The article with the highest score is then used to augment the thesaurus concept. It is the source for the extended gloss, explaining the concept's meaning, synonymous expressions that can be used as additional non-descriptors in the thesaurus, translations of the concept into other languages, and new domain-relevant concepts. Copyright is held by the author/owner(s).
Huang, A, Milne, D, Frank, E & Witten, IH 2008, 'Clustering Documents with Active Learning using Wikipedia', ICDM 2008: EIGHTH IEEE INTERNATIONAL CONFERENCE ON DATA MINING, PROCEEDINGS, 8th IEEE International Conference on Data Mining, IEEE COMPUTER SOC, Pisa, ITALY, pp. 839-844.View/Download from: Publisher's site
Medelyan, O, Witten, IH & Milne, D 2008, 'Topic indexing with wikipedia', AAAI Workshop - Technical Report, pp. 19-24.
Wikipedia can be utilized as a controlled vocabulary for identifying the main topics in a document, with article titles serving as index terms and redirect titles as their synonyms. Wikipedia contains over 4M such titles covering the terminology of nearly any document collection. This permits controlled indexing in the absence of manually created vocabularies. We combine state-of-the-art strategies for automatic controlled indexing with Wikipedia's unique property-a richly hyperlinked encyclopedia. We evaluate the scheme by comparing automatically assigned topics with those chosen manually by human indexers. Analysis of indexing consistency shows that our algorithm performs as well as the average person.
Milne, D & Witten, IH 2008, 'An effective, low-cost measure of semantic relatedness obtained from wikipedia links', AAAI Workshop - Technical Report, pp. 25-30.
This paper describes a new technique for obtaining measures of semantic relatedness. Like other recent approaches, it uses Wikipedia to provide structured world knowledge about the terms of interest. Our approach is unique in that it does so using the hyperlink structure of Wikipedia rather than its category hierarchy or textual content. Evaluation with manually defined measures of semantic relatedness reveals this to be an effective compromise between the ease of computation of the former approach and the accuracy of the latter. Copyright © 2008.
This paper describes how to automatically cross-reference documents with Wikipedia: the largest knowledge base ever known. It explains how machine learning can be used to identify significant terms within unstructured text, and enrich it with links to the appropriate Wikipedia articles. The resulting link detector and disambiguator performs very well, with recall and precision of almost 75%. This performance is constant whether the system is evaluated on Wikipedia articles or -real world- documents. This work has implications far beyond enriching documents with explanatory links. It can provide structured knowledge about any unstructured fragment of text. Any task that is currently addressed with bags of words-indexing, clustering, retrieval, and summarization to name a few-could use the techniques described here to draw on a vast network of concepts and semantics. © 2008 ACM.
Milne, DN, Nichols, DM & Witten, IH 2008, 'A competitive environment for exploratory query expansion', Proceedings of the ACM International Conference on Digital Libraries, pp. 197-200.View/Download from: Publisher's site
Most information workers query digital libraries many times a day. Yet people have little opportunity to hone their skills in a controlled environment, or compare their performance with others in an objective way. Conversely, although search engine logs record how users evolve queries, they lack crucial information about the user's intent. This paper describes an environment for exploratory query expansion that pits users against each other and lets them compete, and practice, in their own time and on their own workstation. The system captures query evolution behavior on predetermined information-seeking tasks. It is publicly available, and the code is open source so that others can set up their own competitive environments. Copyright 2008 ACM.
Milne, D, Witten, IH & Nichols, DM 2007, 'A knowledge-based search engine powered by Wikipedia', International Conference on Information and Knowledge Management, Proceedings, pp. 445-454.View/Download from: Publisher's site
This paper describes Koru, a new search interface that offers effective domain-independent knowledge-based information retrieval. Koru exhibits an understanding of the topics of both queries and documents. This allows it to (a) expand queries automatically and (b) help guide the user as they evolve their queries interactively. Its understanding is mined from the vast investment of manual effort and judgment that is Wikipedia. We show how this open, constantly evolving encyclopedia can yield inexpensive knowledge structures that are specifically tailored to expose the topics, terminology and semantics of individual document collections. We conducted a detailed user study with 12 participants and 10 topics from the 2005 TREC HARD track, and found that Koru and its underlying knowledge base offers significant advantages over traditional keyword search. It was capable of lending assistance to almost every query issued to it; making their entry more efficient, improving the relevance of the documents they return, and narrowing the gap between expert and novice seekers. Copyright 2007 ACM.
Milne, D 2007, 'Computing semantic relatedness using Wikipedia Link structure', Proceedings of NZCSRSC 2007, the 5th New Zealand Computer Science Research Student Conference.
This paper describes a new technique for obtaining measures of semantic relatedness. Like other recent approaches, it uses Wikipedia to provide a vast amount of structured world knowledge about the terms of interest. Our system, the Wikipedia Link Vector Model or WLVM, is unique in that it does so using only the hyperlink structure of Wikipedia rather than its full textual content. To evaluate the algorithm we use a large, widely used test set of manually defined measures of semantic relatedness as our bench-mark. This allows direct comparison of our system with other similar techniques.
Milne, DN 2007, 'Exploiting web 2.0 forallknowledge-based information retrieval', International Conference on Information and Knowledge Management, Proceedings, pp. 69-76.View/Download from: Publisher's site
This paper describes ongoing research into obtaining and using knowledge bases to assist information retrieval. These structures are prohibitively expensive to obtain manually, yet automatic approaches have been researched for decades with limited success. This research investigates a potential shortcut: a way to provide knowledge bases automatically, without expecting computers to replace expert human indexers. Instead we aim to replace the professionals with thousands or even millions of amateurs: with the growing community of contributors who form the core of Web 2.0. Specifically we focus on Wikipedia, which represents a rich tapestry of topics and semantics and a huge investment of human effort and judgment. We show how this can be directly exploited to provide manually-defined yet inexpensive knowledge-bases that are specifically tailored to expose the topics, terminology and semantics of individual document collections. We are also concerned with how best to make these structures available to users, and aim to produce a complete knowledge-based retrieval system-both the knowledge base and the tools to apply it-that can be evaluated by how well it assists real users in performing realistic and practical information retrieval tasks. To this end we have developed Koru, a new search engine that offers concrete evidence of the effectiveness of our Web 2.0 based techniques for assisting information retrieval. © 2007 ACM.
Milne, D, Medelyan, O & Witten, IH 2007, 'Mining domain-specific thesauri from Wikipedia: A case study', Proceedings - 2006 IEEE/WIC/ACM International Conference on Web Intelligence (WI 2006 Main Conference Proceedings), WI'06, pp. 442-448.View/Download from: Publisher's site
Domain-specific thesauri are high-cost, high-maintenance, high-value knowledge structures. We show how the classic thesaurus structure of terms and links can be mined automatically from Wikipedia. In a comparison with a professional thesaurus for agriculture we find that Wikipedia contains a substantial proportion of its concepts and semantic relations; furthermore it has impressive coverage of contemporary documents in the domain. Thesauri derived using our techniques capitalize on existing public efforts and tend to reflect contemporary language usage better than their costly, painstakingly-constructed manual counterparts. © 2006 IEEE.