Factors That Affect The Combined Analysis Of Sa And Iud

In this section, we present a four-step methodology to quantify distinct factors that affect the combined analysis of SA and IUD in real domains, as well as the potential benefits of this type of analysis. Basically, the proposed methodology takes into account the domain s temporal dynamics, the sampling sensitivity of the methods and the observed reciprocity between the collective opinion and opinions propagated by opinion-leaders. Through this methodology, we intend to quantify some important issues related to combine SA and IUD. We are not assuming a closed and complete assessment on all existing issues, which comprise promising research directions for the area.looseness=-1 subsection{Temporal Dynamics Analysis} A main†¦show more content†¦First, we derive the collective opinion $O$ of a whole data sample $D$, using a SA method existing in the literature. Specifically, we adopt in this work the method SACI cite{jws2015}. SACI is relevant to our goal since it was originally proposed for estimating efficiently collective sentiment on data samples, instead of aggregating the sentiment derived for each individual document. Further, the authors demonstrated that SACI is more effective in estimating the collective opinion than aggregation-based SA methods. SACI represents $O$ as a distribution probability among the sentiment classes positive, negative and neutral. Thus, we split $D$ into time units of equal size (e.g., days, weeks, months). Then, we estimate the collective opinion $O_t$ using only the posts belonging to each distinct time unit $t$. Finally, we perform a visual inspection on the derived distributions. The more dynamic a domain, the more different are opinions estimated on distinct time units.looseness=-1 In turn, we measure drifts on the subset of opinion-leaders over time as follows. First, we identify the ordered list $L$ of top-k opinion-leaders on $D$ by using an IUD method. Specifically, we use the presented in cite{iccs2015}, a meta-learning strategy based on PCA that combines linearly orthogonal information exploited by distinct state-of-the-art IUD methods. We will call this method as PCA-IUD. We chose PCA-IUD since it combines