Smartphone Message Sentiment Analysis

Abstract

Humans tend to use specific words to express their emotional states in written and oral communications. Scientists in the area of text mining and natural language processing have studied sentiment fingerprints residing in text to extract the emotional polarity of customers for a product or to evaluate the popularity of politicians. Recent research focused on micro-blogging has found notable similarities between Twitter feeds and SMS (short message service) text messages. This paper investigates the common characteristics of both formats for sentiment analysis purposes and verifies the correctness of the similarity assumption. A lexicon-based approach is used to extract and compute the sentiment scores of SMS messages found on smartphones. The data is presented along a timeline that depicts a sender's emotional fingerprint. This form of analysis and visualization can enrich a forensic investigation by conveying potential psychological patterns from text messages.

Optimizing Short Message Text Sentiment Analysis for Mobile Device Forensics

“…This can be explained by the use of a different machine learning algorithm in this research and the introduction of novel pre-processing techniques. The test set used in this research was the same as that used in [3], where an F-score of 68.8% was obtained. Based on the F-score of 73.59% obtained in this work, it can be deduced that the current classifier achieved a percentage increase of 6.96%.…”

“…Sentiment analysis techniques include: (i) lexicon-based methods [3]; (ii) machine learning methods [1]; and (iii) hybrid approaches that combine lexicon-based and machine learning methods [1,10]. When treating sentiment analysis as a classification task, machine learning algorithms that are known to perform well in text classification are often used.…”

“…Andriotis et al [3] have applied sentiment analysis to augment digital forensic investigations by retrieving opinion information from SMS texts found in mobile devices. A lexicon-based technique was used for sentiment polarity classification and a proof-of-concept system was developed to visualize mood patterns extracted from SMS message databases.…”

“…The training dataset contained 8,120 tweets (positive: 37.2%, negative: 14.7% and neutral: 48.1%). The testing dataset from [3] was also employed, making it possible to compare the results directly.…”

“…In the research literature, applications of sentiment analysis have been proposed in a variety of fields, especially related to social media and micro blogging services. As discussed in [1,3], sentiment information can also be useful in forensic investigations of smartphones.…”