Background: Twelve lead ECGs are a core diagnostic tool for cardiovascular diseases. Here, we describe and analyse an ensemble deep neural network architecture to classify 24 cardiac abnormalities from 12-lead ECGs. Method: We proposed a squeeze and excite ResNet to automatically learn deep features from 12-lead ECGs, in order to identify 24 cardiac conditions. The deep features were augmented with age and gender features in the final fully connected layers. Output thresholds for each class were set using a constrained grid search. To determine why the model made incorrect predictions, two expert clinicians independently interpreted a random set of 100 misclassified ECGs concerning Left Axis Deviation. Results: Using the bespoke weighted accuracy metric, we achieved a 5-fold cross validation score of 0.684, and sensitivity and specificity of 0.758 and 0.969, respectively. We scored 0.520 on the full test data, and ranked 2nd out of 41 in the official challenge rankings. On a random set of misclassified ECGs, agreement between two clinicians and training labels was poor (clinician 1: kappa = -0.057, clinician 2: kappa = -0.159). In contrast, agreement between the clinicians was very high (kappa = 0.92). Discussion: The proposed prediction model performed well on the validation and hidden test data in comparison to models trained on the same data. We also discovered considerable inconsistency in training labels, which is likely to hinder development of more accurate models.

Link prediction in signed social networks is challenging because of the existence and imbalance of the three kinds of social status (positive, negative and no-relation). Furthermore, there are a variety types of no-relation status in reality, e.g., strangers and frenemies, which cannot be well distinguished from the other linked status by existing approaches. In this paper, we propose a novel Framework of Integrating both Latent and Explicit features (FILE), to better deal with the no-relation status and improve the overall link prediction performance in signed networks. In particular, we design two latent features from latent space and two explicit features by extending social theories, and learn these features for each user via matrix factorization with a specially designed ranking-oriented loss function. Experimental results demonstrate the superior of our approach over state-of-the-art methods.

We rethink the link prediction problem in signed social networks by also considering "no-relation" as a future status of a node pair, rather than simply distinguishing positive and negative links proposed in the literature. To understand the underlying mechanism of link formation in signed networks, we propose a feature framework on the basis of a thorough exploration of potential features for the newly identified problem. Grounded on the framework, we also design a trinary classification model, and experimental results show that our method outperforms the state-of-the-art approaches.

Reputation models depend on the ratings provided by buyers togauge the reliability of sellers in multi-agent based e-commerce environment. However, there is no prevention forthe cases in which a buyer misjudges a seller, and provides a negative rating to an original satisfactory transaction. In this case,how should the seller get his reputation repaired andutility loss recovered? In this work, we propose a mechanism to mitigate the negativeeffect of the misreported ratings. It temporarily inflates the reputation of thevictim seller with a certain value for a period of time. This allows the seller to recover hisutility loss due to lost opportunities caused by the misreported ratings. Experiments demonstrate the necessity and effectiveness of the proposed mechanism.

Although users' preference is semantically reflected in the free-form review texts, this wealth of information was not fully exploited for learning recommender models. Specifically, almost all existing recommendation algorithms only exploit rating scores in order to find users' preference, but ignore the review texts accompanied with rating information. In this paper, we propose a novel matrix factorization model (called TopicMF) which simultaneously considers the ratings and accompanied review texts. Experimental results on 22 real-world datasets show the superiority of our model over the state-of-the-art models, demonstrating its effectiveness for recommendation tasks.

Trust has been extensively exploited to improve the predictive The dependency between the trust aspects is captured by accuracy of recommendations by ameliorating the issues a Guassian radial basis kernel function. Then, we incorporate such as data sparsity and cold start that recommender the trust information into the probabilistic matrix factorization systems inherently suffer from (Massa and Avesani 2007; model (Mnih and Salakhutdinov 2007) by modeling Ma et al. 2008). Basically, trust provides additional information trust as jointly conditioning on the trust value obtained from which user preference can be better modeled, alternative from the SVR model, as well as similarity between the corresponding or complementary to rating-based similarity.