Purpose – We analyze streams of microblogs in terms of both contents and distribution network to create a model of people behavior to be used as an indication for future similar events. We present our methodology and some preliminary results of its application to new movie launches through the analysis of about 2 million downloaded in the US in 72 hours around the events in a 4 months period. Design/methodology/approach – We use a combination of semantic and topological analyses that demonstrated good results identifying conversational patterns to be used not only to assess how people feel about the event but also to develop a better understanding about what people say and how people talk about the event. To obtain the what people say, the methodology and the tools automatically analyze the conversations, with the support of visualization components. The how people talk is addressed detecting patterns in the conversations and evaluating the correlation of those patterns with performance indicators. Structure and contents are analyzed through a set of semantic and topological metrics to assess the content generated in different conversations. Metrics are then collected in a single dataset, which is later used to train and test a data mining model. Originality/value – Several commercial platforms are available to retrieve and assess to some extent this collective judgment through sentiment analysis. Sentiment, however, just measures how the “crowd” feels about a product, but does not offer insight on the structure and the determinants of the customers’ preferences nor provides indications about future behavior. Our challenge is instead to dig deeper into Twitter streams to capture and assess structured contents that are embedded in them through an analytical and quantitative approach. Practical implications – The elicitation of organized content from Twitter streams could support market analysts to achieve a better understanding of consumers’ perceptions and to better manage the reach-richness trade off between qualitative and quantitative market analyses. An additional practical use of our study can be to find a way to ascertain the presence of given semantic patterns in Twitter streams that can be predictive of early market success for a new product.