This presentation covers three main issues. In the first part I present my research work on sentiment analysis using neural network architectures that I developed. The second part briefly introduces extractive and abstractive text summarization and their main applications. In the context of ELITER project, the final minuting phase requires summarization of meeting dialogs. Sentence compression will be followed by summarization of entire meeting transcript using neural networks. One problem here is the need to carefully preprocess meeting dialog texts which are peculiar and contain frequent disfluencies. Various segmentation functions will be explored for this purpose. A possible extension could be a third summarization module for generating decision summaries of each meeting transcript. The final part of the presentation introduces research opportunities on sentiment analysis in Czech. Combining rich linguistic features with neural network structures could further enhance the accuracy of sentiment identification in Czech texts.