Josiah Poon

A neural network based approach to automated e-mail classification

We present a neural network based system for automated e-mail filing into folders and anti-spam filtering. The experiments show that it is more accurate than several other techniques. We also investigate the effects of various feature selection, weighting and normalization methods, and also the portability of the anti-spam filter across different users.

Learning to classify e-mail

In this paper we study supervised and semi-supervised classification of e-mails. We consider two tasks: filing e-mails into folders and spam e-mail filtering. Firstly, in a supervised learning setting, we investigate the use of random forest for automatic e-mail filing into folders and spam e-mail filtering. We show that random forest is a good choice for these tasks as it runs fast on large and high dimensional databases, is easy to tune and is highly accurate, outperforming popular algorithms such as decision trees, support vector machines and naive Bayes. We introduce a new accurate feature selector with linear time complexity. Secondly, we examine the applicability of the semi-supervised co-training paradigm for spam e-mail filtering by employing random forests, support vector machines, decision tree and naive Bayes as base classifiers. The study shows that a classifier trained on a small set of labelled examples can be successfully boosted using unlabelled examples to accuracy rate of only 5% lower than a classifier trained on all labelled examples. We investigate the performance of co-training with one natural feature split and show that in the domain of spam e-mail filtering it can be as competitive as co-training with two natural feature splits.

Co-training with a Single Natural Feature Set Applied to Email Classification

When dealing with information overload from the Internet, such as the classification of Web pages and the filtering of email spam, a new technique called co-training has been shown to be a promising approach to help build more accurate classifiers. Co-training allows classifiers to learn with fewer labelled documents by taking advantage of the more abundant unclassified documents. However, conventional co-training requires the dataset to be described by two disjoint and natural feature sets that are sufficiently redundant. In many practical situations, it is not intuitively obvious how to obtain two natural feature sets. This paper shows that when only a single natural feature set is used, the performance of co-training is beneficial in the application of email classification.

Co-evolution and emergence in design

Evolution as a metaphor borrowed from nature can be used to describe a design process. However, this has generally been applied to the evolution of a solution which assumes the problem does not change throughout the process. This is a naive assumption in design because the problem indeed changes. This paper considers the evolution of both the problem and solution and introduces co-evolutionary design. This paper proposes two approaches to implementing co-evolutionary design and also addresses the related issues of evaluation and termination in a computational model. Finally, the paper considers how a co-evolutionary system can generate and recognize emergent structure and behaviour.

Applying Data Mining to Pseudo-Relevance Feedback for High Performance Text Retrieval

In this paper, we investigate the use of data mining, in particular the text classification and co-training techniques, to identify more relevant passages based on a small set of labeled passages obtained from the blind feedback of a retrieval system. The data mining results are used to expand query terms and to re-estimate some of the parameters used in a probabilistic weighting function. We evaluate the data mining based feedback method on the TREC HARD data set. The results show that data mining can be successfully applied to improve the text retrieval performance. We report our experimental findings in detail.

Evidence-based toxicity evaluation and scheduling of Chinese herbal medicines.

ETHNOPHARMACOLOGICAL RELEVANCE While there is an increasing number of toxicity report cases and toxicological studies on Chinese herbal medicines, the guidelines for toxicity evaluation and scheduling of Chinese herbal medicines are lacking. AIM The aim of this study was to review the current literature on potentially toxic Chinese herbal medicines, and to develop a scheduling platform which will inform an evidence-based regulatory framework for these medicines in the community. MATERIALS AND METHODS The Australian and Chinese regulations were used as a starting point to compile a list of potentially toxic herbs. Systematic literature searches of botanical and pharmaceutical Latin name, English and Chinese names and suspected toxic chemicals were conducted on Medline, PubMed and Chinese CNKI databases. RESULTS Seventy-four Chinese herbal medicines were identified and five of them were selected for detailed study. Preclinical and clinical data were summarised at six levels. Based on the evaluation criteria, which included risk-benefit analysis, severity of toxic effects and clinical and preclinical data, four regulatory classes were proposed: Prohibited for medicinal usage, which are those with high toxicity and can lead to injury or death, e.g., aristolochia; Restricted for medicinal usage, e.g., aconite, asarum, and ephedra; Required warning label, e.g., coltsfoot; and Over-the-counter herbs for those herbs with a safe toxicity profile. CONCLUSION Chinese herbal medicines should be scheduled based on a set of evaluation criteria, to ensure their safe use and to satisfy the need for access to the herbs. The current Chinese and Australian regulation of Chinese herbal medicines should be updated to restrict the access of some potentially toxic herbs to Chinese medicine practitioners who are qualified through registration.

Formalising Design Exploration as Co-Evolution

This paper introduces a model for design exploration based on notions of evolution and demonstrates computational co-evolution using a modified genetic algorithm (GA). Evolution is extended to consider co-evolution where two systems evolve in response to each other. Co-evolution in design exploration supports the change, over time, of the design solution and the design requirements. The basic GA, which does not support our exploration model, evaluates individuals from a population of design solutions with an unchanged fitness function. This approach to evaluation implements search with a prefixed goal. Modifications to the basic GA are required to support exploration. Two approaches to implement a co-evolving GA are: a combined gene approach and a separate spaces approach. The combined gene approach includes the representation of the requirements and the solution within the genotype. The separate spaces approach models the requirements and the solutions as separately evolving interacting populations of genotypes. The combined gene approach is developed further in this paper and used to demonstrate design exploration in the domain of braced frame design for buildings. The issues related to the coding of the genotype, mapping to a phenotype, and evaluation of the phenotype are addressed. Preliminary results of co-evolution are presented that show how exploration differs from search.

INTIMATE: a Web-based movie recommender using text categorization

We present INTIMATE, a Web-based movie recommender that makes suggestions by using text categorization to learn from movie synopses. The performance of various feature representations, feature selectors, feature weighting mechanisms and classifiers is evaluated and discussed. INTIMATE was also compared with a feature-based movie recommender. The results show that the text-based approach outperforms the feature-based if the ratio of the number of user ratings to the vocabulary size is high.

A novel approach in discovering significant interactions from TCM patient prescription data

The efficacy of a traditional Chinese medicine medication derives from the complex interactions of herbs or Chinese Materia Medica in a formula. The aim of this paper is to propose a new approach to systematically generate combinations of interacting herbs that might lead to good outcome. Our approach was tested on a data set of prescriptions for diabetic patients to verify the effectiveness of detected combinations of herbs. This approach is able to detect effective higher orders of herb-herb interactions with statistical validation. We present an exploratory analysis of clinical records using a pattern mining approach called Interaction Rules Mining.

Novel two-stage analytic approach in extraction of strong herb-herb interactions in TCM clinical treatment of insomnia

In this paper, we aim to investigate strong herb-herb interactions in TCM for effective treatment of insomnia. Given that extraction of herb interactions is quite similar to gene epistasis study due to non-linear interactions among their study factors, we propose to apply Multifactor Dimensionality Reduction (MDR) that has shown useful in discovering hidden interaction patterns in biomedical domains. However, MDR suffers from high computational overhead incurred in its exhaustive enumeration of factors combinations in its processing. To address this drawback, we introduce a two-stage analytical approach which first uses hierarchical core sub-network analysis to pre-select the subset of herbs that have high probability in participating in herb-herb interactions, which is followed by applying MDR to detect strong attribute interactions in the pre-selected subset. Experimental evaluation confirms that this approach is able to detect effective high order herb-herb interaction models in high dimensional TCM insomnia dataset that also has high predictive accuracies.