Although the neoclassical finance paradigm of efficient markets provides the proposition that stock returns are unpredictable (Fama, 1970), a large body of contradictory empirical evidence has brought this theory into question (Baker and Wurgler, 2000; Cochrane, 2000). In light of this evidence, behavioral finance has been proposed as an alternative theoretical paradigm to explain stock returns. The key implication of behavioral finance is that the emotions and moods of investors play an important role in financial decisions (Nofsinger, 2005). Moreover, the presence of irrationality and the emotive basis of decisions made by noise-traders, who comprise a relatively large proportion of stock market participants (Black, 1986), has resulted in investor sentiment being considered to influence investor decisionmaking, and hence stock returns. This new paradigm of stock market behavior has resulted in the need to develop an accurate measure of investor sentiment (Chan and Chong, 2017). Despite a large number of studies proposing a relationship between investor sentiment extracted from social media and stock market returns, there is no consensus in empirical studies whether this theoretical relationship is supported in the data. Proponents of behavioral finance argue that this lack of empirical evidence can be attributed mainly to problems with the measurement of investor sentiment through social networks in existing studies of financial markets. These problems include: the absence of an accurate approach for measuring investor sentiment (Renault, 2017; Oh and Sheng, 2011); use of datasets from platforms that do not accurately represent investors (Bollen et al., 2011; Ranco et al., 2015); and the use of short sample periods (Bollen et al., 2011; Li et al., 2018). Motivated by these problems, this study applies recent advances in the domain-general sentiment analysis literature to data from a finance-related social media platform to construct a more accurate decision support system in the context of investor sentiment classification. The collection of investor sentiment data from Internet-based microblogs overcomes issues that have been identified from the use of questionnaires, such as errors due to impaired questionnaire design (Brace, 2008) and inaccurate or untruthful participant responses (Singer, 2002).