@article { , title = {Optimizing short message text sentiment analysis for mobile device forensics}, abstract = {© IFIP International Federation for Information Processing 2016. Mobile devices are now the dominant medium for communications. Humans express various emotions when communicating with others and these communications can be analyzed to deduce their emotional inclinations. Natural language processing techniques have been used to analyze sentiment in text. However, most research involving sentiment analysis in the short message domain (SMS and Twitter) do not account for the presence of non-dictionary words. This chapter investigates the problem of sentiment analysis in short messages and the analysis of emotional swings of an individual over time. This provides an additional layer of information for forensic analysts when investigating suspects. The maximum entropy algorithm is used to classify short messages as positive, negative or neutral. Non-dictionary words are normalized and the impact of normalization and other features on classification is evaluated; in fact, this approach enhances the classification F-score compared with previous work. A forensic tool with an intuitive user interface has been developed to support the extraction and visualization of sentiment information pertaining to persons of interest. In particular, the tool presents an improved approach for identifying mood swings based on short messages sent by subjects. The timeline view provided by the tool helps pinpoint periods of emotional instability that may require further investigation. Additionally, the Apache Solr system used for indexing ensures that a forensic analyst can retrieve the desired information rapidly and efficiently using faceted search queries.}, doi = {10.1007/978-3-319-46279-0\_4}, isbn = {9783319462783}, issn = {1868-4238}, journal = {IFIP Advances in Information and Communication Technology}, pages = {69-87}, publicationstatus = {Published}, publisher = {Springer Verlag (Germany)}, url = {https://uwe-repository.worktribe.com/output/916108}, volume = {484}, keyword = {sentiment analysis, text mining, SMS, Twitter, normalization}, year = {2016}, author = {Aboluwarin, Oluwapelumi and Andriotis, Panagiotis and Takasu, Atsuhiro and Tryfonas, Theo} editor = {Peterson, Gilbert and Shenoi, Sujeet} }