Search

A novel binary decision architecture (BDA) for broadcast news audio classification task is presented in this paper. The idea of developing such architecture came from the fact that the appropriate combination of multiple binary classifiers for two-class discrimination problem can reduce a miss-classification error without rapid increase in computational complexity. The core element of classification architecture is represented by a binary decision (BD) algorithm that performs discrimination between each pair of acoustic classes, utilizing two types of decision functions. The first one is represented by a simple rule-based approach in which the final decision is made according to the value of selected discrimination parameter. The main advantage of this solution is relatively low processing time needed for classification of all acoustic classes. The cost for that is low classification accuracy. The second one employs support vector machine (SVM) classifier. In this case, the overall classification accuracy is conditioned by finding the optimal parameters for decision function resulting in higher computational complexity and better classification performance. The final form of proposed BDA is created by combining four BD discriminators supplemented by decision table. The effectiveness of proposed BDA, utilizing rule-based approach and the SVM classifier, is compared with two most popular strategies for multiclass classification, namely the binary decision trees (BDT) and the One-Against-One SVM (OAOSVM). Experimental results show that the proposed classification architecture can decrease the overall classification error in comparison with the BDT architecture. On the contrary, an optimization technique for selecting the optimal set of training data is needed in order to overcome the OAOSVM.

We model the decision making process used by Experts at the Canadian Innovation Centre to classify early stage venture proposals based on potential commercial success. The decision is based on thirty-seven attributes that take values in { - 1 , 0 , 1 } . We adopt a conjunctive decision framework due to Astebro and Elhedhli (2005) that selects a subset of attributes and determines two threshold values: one for the maximum allowed negatives (n) and one for minimum required positives (p). A proposal is classified as a success if the number of positives is greater than or equal to p and the number of negatives is less than or equal to n over the selected attributes. Based on a data set of 561 observations, the selection of attributes and the determination of the threshold values is modeled as a large-scale mixed integer program. Two solution approaches are explored: Benders decomposition and Tabu search. The first, was very slow to converge, while the second provided high quality solutions quickly. Tabu search provides excellent classification accuracy for predicting commercial successes as well as replicating Experts’ forecasts, opening the venue for the use of Tabu search in scoring and classification problems.

The trajectories of social processes (e.g., peer pressure, imitation, and assimilation) that take place on social networks depend on the structure of those networks. Thus, to understand a social process or to predict the associated outcomes accurately, marketers would need good knowledge of the social network structure. However, many social networks of relevance to marketers are large, complex, or hidden, making it prohibitively expensive to map out an entire social network. Instead, marketers often need to work with a sample (i.e., a subgraph) of a social network. In this paper we evaluate the efficacy of nine different sampling methods for generating subgraphs that recover four structural characteristics of importance to marketers, namely, the distributions of degree, clustering coefficient, betweenness centrality, and closeness centrality, which are important for understanding how social network structure influences outcomes of processes that take place on the network. Via extensive simulations, we find that sampling methods differ substantially in their ability to recover network characteristics. Traditional sampling procedures, such as random node sampling, result in poor subgraphs. When the focus is on understanding local network effects (e.g., peer influence) then forest fire sampling with a medium burn rate performs the best, i.e., it is most effective for recovering the distributions of degree and clustering coefficient. When the focus is on global network effects (e.g., speed of diffusion, identifying influential nodes, or the “multiplier” effects of network seeding), then random-walk sampling (i.e., forest-fire sampling with a low burn rate) performs the best, and it is most effective for recovering the distributions of betweenness and closeness centrality. Further, we show that accurate recovery of social network structure in a sample is important for inferring the properties of a network process, when one observes only the process in the sampled network. We validate our findings on four different real-world networks, including a Facebook network and a co-authorship network, and conclude with recommendations for practice.