Online communication has increased the need to interpret complex emotions rapidly; due to the volatility of the data involved, machine learning tasks that process text aim to address the related challenges. Exploring text in comments that supports ideas through computational methods is a logical next step considering similar research for these question-and-answer sites. A lack of current algorithms that can accurately predict accepted answers with equal votes suggests a gap in this knowledge. Measuring collaborative signals in comments finds common keywords to move a problem toward a solution.Using a dataset from questions posted to Stack Exchange on the subject area of machine learning, the researchers constructed a model using the comments of posts made on accepted answers. Intending to discover whether predictions of marked solutions are accurate by treating the comments as reviews, the researchers find a reduction in error by incorporating reviews of answers as a feature in the predictive algorithm.

Exploring help speech through computational methods is a counterbalance to similar research for offensive speech. A lack of agreement in sentiment dictionaries around the value of help speech suggests a gap in this research. Using HathiTrust, an online library that exposes digital copies of copyrighted works for token-based analysis, the researchers explore help speech in eighty-seven (87) artifacts authored by the Dalai Lama. This work confirms that the sentiment of peace artifacts is generally positive and suggests further work to explore where help speech and offensive speech may overlap.

With the continued explosion of cyber traffic across billions of IP addresses around the globe, it has become extremely challenging to analyze networks due to their growing size and complexity. One promising solution is to identify the communities in the network structure and perform analysis at a community level rather than at a network node level. However, beyond simply detecting communities, it is also crucial to evaluate the quality of the detected communities; modularity is one such metric for evaluation. When evaluating the modularity index, researchers have considered null models for graphs with specific structural characteristics. However, most real-world complex networks as a whole do not exhibit one specific characteristic but instead consist of various identifiable subgraphs that do respectively exhibit particular characteristics, and accordingly, formulating a null model for these individual subgraphs may improve the modularity value and thereby improve the quality of the partitions, i.e., the detected communities. This paper investigates the extent to which the modularity value increases when a bipartite subgraph is taken into consideration while performing community detection. Our empirical results suggest that the quality of the detected communities is enhanced by leveraging the presence of bipartite subnetworks in complex networks.

There are many challenges in digital forensic investigations involving large-scale computer networks; these include large volume of data, the limited scope of tools, the financial burdens of purchasing and licensing those tools, and identifying salient evidence from the vast amounts of network data. We have implemented a collection of open-source tools and code wrappers to provide a tool for network forensic investigators to capture, selectively analyze, and reconstruct files from network traffic. The main functions of this tool (FileTSAR) are capturing data flows and providing a mechanism to selectively reconstruct documents, images, email, and VoIP conversations. To validate the large-scale capabilities of the toolkit, we conducted a "stress test" of the system using approximately 123,500,000 packets from a collection of packet capture files totaling nearly 100GB. Additionally, sixteen (16) digital forensic examiners participated in a 3-day law enforcement training workshop for FileTSAR from across the United States; the examiners expressed substantial support for FileTSAR with large-scale investigations as well as an interest in a scaled-down version for smaller agencies with storage, budget, and back-end support limitations.