Game reviews function as an important customer-created resource for game studies as they allow practitioners and developers to analyze players' opinions. Despite this, there are few studies that undertake comparative evaluations of topic modeling approaches in the context of video game data analysis or assess the results' practical efficacy. Accordingly, this article aims to evaluate the performance of three topic modeling algorithms—latent Dirichlet allocation, non-negative matrix factorization, and BERTopic—as utilized within game reviews study and further to examine the results' reception within the video game industry. This study first uses the gameNo Man's Skyas a case study to evaluate the performance of different models in the same game context. According to our experiments based on Steam game reviews, the topic's Uci coherence score as identified by the BERTopic model can reach 0.279, which is higher than the other two models, with the extracted keywords allowing humans to interpret the themes when mapping them to original reviews. Semi-structured interviews with seven developers are then presented, which demonstrate that the information we provided is useful to improve their games and track players' opinions.

Ensemble clustering is a technique which combines multiple clustering results, and instance weighting is a technique which highlights important instances in a dataset.Both techniques are known to enhance clustering performance and robustness.In this research, ensembles and instance weighting are integrated with the spectral clustering algorithm.We believe this is the first attempt at creating diversity in the generative mechanism using density based instance weighting for a spectral ensemble.The proposed approach is empirically validated using synthetic datasets comparing against spectral and a spectral ensemble with random instance weighting.Results show that using the instance weighted sub-sampling approach as the generative mechanism for an ensemble of spectral clustering leads to improved clustering performance on datasets with imbalanced clusters.

Creating speech emotion recognition models com-parable to the capability of how humans recognise emotions is a long-standing challenge in the field of speech technology with many potential commercial applications. As transformer-based architectures have recently become the state-of-the-art for many natural language processing related applications, this paper investigates their suitability for acoustic emotion recognition and compares them to the well-known AlexNet convolutional approach. This comparison is made using several publicly available speech emotion corpora. Experimental results demonstrate the efficacy of the different architectural approaches for particular emotions. The results show that the transformer-based models outperform their convolutional counterparts yielding F1-scores in the range [70.33%, 75.76%]. This paper further provides insights via dimensionality reduction analysis of output layer activations in both architectures and reveals significantly improved clustering in transformer-based models whilst highlighting the nuances with regard to the separability of different emotion classes.

Sentiment Analysis is a popular classification problem where the degree of emotion is analysed based on text. Convolution Neural Network (CNN) are part of various sentiment classification models, but their use has been limited to one-dimensional (1D) convolution or shallow two-dimensional (2D) convolution. The primary focus of this research is to use deep 2D-CNN architectures, inspired by the popular computer vision model ResNet, to replace a shallow 2D-CNN in an existing BLSTM (Bidirectional Long Short Term Memory)-2DCNN model. This research investigates a new method for sentiment analysis which is an amalgam of the practices popular in Natural Language Processing (NLP) as well as Computer Vision, striving to extract opinions from a text by transforming text into images. The text images are formed by transforming the word embeddings matrix into a greyscale image with the help of BLSTM cell. Intensive experiments have been conducted on the Sentiment140 dataset with our novel BLSTM-ResNet model, which contains residual blocks with skip connections. Several BLSTM-ResNet variants were tested to investigate the impact of network depth and dataset size on sentiment detection. Moreover, two sets of residual blocks are designed to form our shallow and deep BLSTM-ResNet models. Our best shallow BLSTM-ResNet models have achieved 4.06% and 3.43% increases in accuracy for dataset sizes 80,000 and 200,000 respectively, compared with the baseline BLSTM-2DCNN model. In addition, an overall improvement is observed on accuracy with every additional residual block in our shallow BLSTM-ResNet model until accuracy saturates, and the same trend has been seen on the impact of dataset size on the performance. Our deep BLSTM-ResNet models show the same positive impact of network depth impact and dataset size on sentiment analysis. Further investigation on the shallow and deep BLSTM-ResNet models shows that deep BLSTM-ResNet outperforms shallow BLSTM-ResNet, in general.

Astrophysicists rely on crowd sourced initiatives to classify galaxies in large surveys. The next generation of telescopes will lead to a revolutionary increase in the amount of unlabelled data available to astrophysicists making crowd sourcing infeasible. To cope with this significant increase in data astrophysicists will need unsupervised techniques. In this paper we show that a model using unsupervised learning can automatically categorise galaxies in Hubble Space Telescope survey images. We discover the optimum parameter combination using a grid search like approach. We also present a detailed analysis of the effect of varying parameters. We analyse Hubble Space Telescope images and evaluate the model's performance for identifying major categories of galaxies and for identifying a type of lensed galaxy. We find that the pixel intensity power spectrum is the most effective at identifying elliptical, spiral and background galaxies whereas the rotationally invariant feature transform is more effective at detecting a type of lensed galaxy that occurs as a result of strong lensing.

Adoption of Cloud Computing is on the rise[1] and many datacenter operators adhere to strict energy efficiency guidelines[2]. In this paper a novel approach to scheduling in a Cloud Computing context is proposed. The algorithm Maxmin Fast Track (MXFT) revises the Max-min algorithm to better support smaller tasks with stricter Service Level Agreements (SLAs), which makes it more relevant to Cloud Computing. MXFT is inspired by queuing in supermarkets, where there is a fast lane for customers with a smaller number of items. The algorithm outperforms Max-min in task execution times and outperforms Min-min in overall makespan. A by-product of investigating this algorithm was the development of simulator called “ScheduleSim”[3] which makes it simpler to prove a scheduling algorithm before committing to a specific scheduling problem in Cloud Computing and therefore might be a useful precursor to experiments using the established simulator CloudSim[4].

Gaussian Process is a Machine Learning technique that has been applied to the analysis of percutaneous absorption of chemicals through human skin. The normal, automatic method of setting the hyperparameters associated with Gaussian Processes may not be suitable for small datasets. In this paper we investigate whether a handcrafted search method of determining these hyperparameters is better for such datasets.

Predicting the rate of percutaneous absorption across mammalian and artificial membranes is a complex problem. In previous studies, prediction and accuracy are approached using different machine learning models. Results show that Gaussian processes provided the best result, based on a range of statistical measures. In general the ultimate aim of these machine learning experiments is to try to understand, analyze and predict the percutaneous absorption of drugs across human skin. One way to do this is to select the best set of chemical descriptors and the dataset of synthetic (Polydimethyl siloxane, PDMS) membranes, containing so many descriptors, is considered a suitable dataset to use in this study. Hence, one of the main purposes of the study is to use feature selection methods to select the molecular properties that exert the most important influence on percutaneous absorption across PDMS membranes, in the hope that this will better inform studies on human skin.

We investigate the relationship between faults and five of Fowler et al.'s least-studied smells in code: Data Clumps, Switch Statements, Speculative Generality, Message Chains, and Middle Man. We developed a tool to detect these five smells in three open-source systems: Eclipse, ArgoUML, and Apache Commons. We collected fault data from the change and fault repositories of each system. We built Negative Binomial regression models to analyse the relationships between smells and faults and report the McFadden effect size of those relationships. Our results suggest that Switch Statements had no effect on faults in any of the three systems; Message Chains increased faults in two systems; Message Chains which occurred in larger files reduced faults; Data Clumps reduced faults in Apache and Eclipse but increased faults in ArgoUML; Middle Man reduced faults only in ArgoUML, and Speculative Generality reduced faults only in Eclipse. File size alone affects faults in some systems but not in all systems. Where smells did significantly affect faults, the size of that effect was small (always under 10 percent). Our findings suggest that some smells do indicate fault-prone code in some circumstances but that the effect that these smells have on faults is small. Our findings also show that smells have different effects on different systems. We conclude that arbitrary refactoring is unlikely to significantly reduce fault-proneness and in some cases may increase fault-proneness.

Background: The NASA Metrics Data Program data sets have been heavily used in software defect prediction experiments. Aim: To demonstrate and explain why these data sets require significant pre-processing in order to be suitable for defect prediction. Method: A meticulously documented data cleansing process involving all 13 of the original NASA data sets. Results: Post our novel data cleansing process; each of the data sets had between 6 to 90 percent less of their original number of recorded values. Conclusions: One: Researchers need to analyse the data that forms the basis of their findings in the context of how it will be used. Two: Defect prediction data sets could benefit from lower level code metrics in addition to those more commonly used, as these will help to distinguish modules, reducing the likelihood of repeated data points. Three: The bulk of defect prediction experiments based on the NASA Metrics Data Program data sets may have led to erroneous findings. This is mainly due to repeated data points potentially causing substantial amounts of training and testing data to be identical.

Background: There has been much discussion amongst auto-mated software defect prediction researchers regarding use of the precision and false positive rate classifier performance metrics. Aim: To demonstrate and explain why failing to report precision when using data with highly imbalanced class distributions may provide an overly optimistic view of classifier performance. Method: Well documented examples of how dependent class distribution affects the suitability of performance measures. Conclusions: When using data where the minority class repre-sents less than around 5 to 10 percent of data points in total, failing to report precision may be a critical mistake. Furthermore, deriving the precision values omitted from studies can reveal valuable insight into true classifier performance. I.

Finding the location of binding sites in DNA is a difficult problem. Although the location of some binding sites have been experimentally identified, other parts of the genome may or may not contain binding sites. This poses problems with negative data in a trainable classifier. Here we show that using randomized negative data gives a large boost in classifier performance when compared to the original labeled data.

Many studies have been carried out to predict the presence of software code defects using static code metrics. Such studies typically report how a classifier performs with real world data, but usually no analysis of the predictions is carried out. An analysis of this kind may be worthwhile as it can illuminate the motivation behind the predictions and the severity of the misclassifications. This investigation involves a manual analysis of the predictions made by Support Vector Machine classifiers using data from the NASA Metrics Data Program repository. The findings show that the predictions are generally well motivated and that the classifiers were, on average, more “confident” in the predictions they made which were correct.

Predicting the rate at which a substance will pass through human skin and into the bloodstream is a problem of current interest. We use Gaussian Process modeling to train a set of predictors using every combination of six molecular features. We find that only three of the features are needed for our best predictor. This result could be useful in the further analysis of skin permeability.

Improving predictions of the skin permeability coefficient is a difficult problem. It is also an important issue with the increasing use of skin patches as a means of drug delivery. In this work, we apply K-nearest-neighbour regression, single layer networks, mixture of experts and Gaussian processes to predict the permeability coefficient. We obtain a considerable improvement over the quantitative structure-activity relationship (QSARs) predictors. We show that using five features, which are molecular weight, solubility parameter, lipophilicity, the number of hydrogen bonding acceptor and donor groups, can produce better predictions than the one using only lipophilicity and the molecular weight. The Gaussian process regression with five compound features gives the best performance in this work.

The identification of cis-regulatory binding sites in DNA in multicellular eukaryotes is a particularly difficult problem in computational biology. To obtain a full understanding of the complex machinery embodied in genetic regulatory networks it is necessary to know both the identity of the regulatory transcription factors together with the location of their binding sites in the genome. We show that using an SVM together with data sampling, to integrate the results of individual algorithms specialised for the prediction of binding site locations, can produce significant improvements upon the original algorithms applied to the mouse genome. These results make more tractable the expensive experimental procedure of actually verifying the predictions.

Currently the best algorithms for transcription factor binding site prediction are severely limited in accuracy. There is good reason to believe that predictions from these different classes of algorithms could he used in conjunction to improve the quality of predictions. In this paper, we apply single layer networks, rules sets and support vector machines on predictions from 12 key real valued algorithms. Furthermore, we use a window of consecutive results in the input vector in order to contextualise the neighbouring results. We improve the classification result with the aid of under- and over- sampling techniques. We find that support vector machines outperform each of the original individual algorithms and the other classifiers employed in this work. In particular they have a better tradeoff between recall and precision.

Recently, we have developed the hierarchical generative topographic mapping (HGTM), an interactive method for visualization of large high-dimensional real-valued data sets. We propose a more general visualization system by extending HGTM in three ways, which allows the user to visualize a wider range of data sets and better support the model development process. 1) We integrate HGTM with noise models from the exponential family of distributions. The basic building block is the latent trait model (LTM). This enables us to visualize data of inherently discrete nature, e.g., collections of documents, in a hierarchical manner. 2) We give the user a choice of initializing the child plots of the current plot in either interactive, or automatic mode. In the interactive mode, the user selects "regions of interest", whereas in the automatic mode, an unsupervised minimum message length (MML)-inspired construction of a mixture of LTMs is employed. The unsupervised construction is particularly useful when high-level plots are covered with dense clusters of highly overlapping data projections, making it difficult to use the interactive mode. Such a situation often arises when visualizing large data sets. 3) We derive general formulas for magnification factors in latent trait models. Magnification factors are a useful tool to improve our understanding of the visualization plots, since they can highlight the boundaries between data clusters. We illustrate our approach on a toy example and evaluate it on three more complex real data sets.

Recent work on extracting features of gaps in handwritten text allows a classification of these gaps into inter-word and intra-word classes using suitable classification techniques. In the previous work, we apply 5 different supervised classification algorithms from the machine learning field on both the original gap dataset and the gap dataset with the best features selected using mutual information. In this paper; we improve the classification result with the aid of a set of feature variables of strokes preceding and following each gap. The best classification result attained suggests that the technique we employ is particularly suitable for digital ink manipulation at the level of words.

Recent work on extracting features of gaps in handwritten text allows a classification into inter-word and intra-word classes using suitable classification techniques. In this paper, we apply 5 different supervised classification algorithms from the machine learning field on both the original dataset and a dataset with the best features selected using mutual information. The classifiers are compared by employing McNemar's test. We find that SVMs and MLPs outperform the other classifiers and that preprocessing to select features works well.

There is a notable interest in extending probabilistic generative modeling principles to accommodate for more complex structured data types. In this paper we develop a generative probabilistic model for visualizing sets of discrete symbolic sequences. The model, a constrained mixture of discrete hidden Markov models, is a generalization of density-based visualization methods previously developed for static data sets. We illustrate our approach on sequences representing web-log data and chorals by J.S. Bach.