Aspect Based Sentiment Analysis (ABSA) systems receive as input a set of texts (e.g.,product reviews or messages from social media) discussing a particular entity (e.g., anew model of a mobile phone). The systems attempt to detect the main (e.g., the mostfrequently discussed) aspects (features) of the entity (e.g., ‘battery’, ‘screen’) and toestimate the average sentiment of the texts per aspect (e.g., how positive or negativethe opinions are on average for each aspect). Although several ABSA systems have been proposed, mostly research prototypes, there is no established task decompositionfor ABSA, nore are there any established evaluation measures for the subtasks ABSA systems are required to perform. This thesis, proposes a new task decomposition for ABSA, which contains three main subtasks: aspect term extraction, aspect term aggregation, and aspect term polarity estimation. The first subtask detects single- and multi-word terms naming aspects of the entity being discussed (e.g., ‘battery’, ‘hard disc’), called aspect terms. The second subtask aggregates (clusters) similar aspect terms (e.g., ‘price’ and ‘cost’, but maybe also ‘design’ and ‘color’), depending on user preferences and other restrictions (e.g., the size of the screen where the results of the ABSA system will be shown). The third subtask estimates the average sentiment per aspect term or cluster of aspect terms. For each one of the above mentioned subtasks, benchmark datasets for different kinds of entities (e.g., laptops, restaurants) were constructed during the work of this thesis. New evaluation measures are introduced for each subtask, arguing that they are more appropriate than previous evaluation measures. For each subtask, the thesis also proposes new methods (or improvements over previous methods), showing experimentally on the constructed benchmark datasets that the new methods (or the improved versions) are better or at least comparable to state of the art ones.