The JDPA Sentiment Corpus for the Automotive Domain Jason S. Kessler and Nicolas Nicolov 1 Introduction The expression of sentiment is a complex phenomenon which is intertwined into the semantic structure of text (Polanyi and Zaenen 2006). A document-level label, such as positive or negative, does not present a full representation of all sentiment present in a document. Sentiment, which we define as evaluation, is expressed to- ward discourse entities by means of individual expressions of sentiment targeted at mentions of those entities. These expressions of sentiment are often rooted in single or multi-word units, whose positive or negativeness may be impacted by the context. Elements in the context that can alter the polarity include negations and terms which can alter the truth-value of an expression of sentiment, as well as less understood phenomena such as sarcasm and tone. While sentiment toward individual mentions of an entity contribute to its overall sentiment, sentiment toward another, related entity such as a part or a feature may also contribute. Sentiment directed toward individual entities can also effect other entities when comparisons among entities are made. An additional dimension of the phenomena is that certain expressions of sentiment may be attributed to discourse participants other than the speaker. Our goal is to annotate structures pertinent to sentiment that can be combined to formally explain the sentiment that occurs in a document. The J.D. Power and Associates (JDPA) Sentiment Corpus consists of user- generated content (blog posts) containing opinions about automobiles. Specifically, Jason S. Kessler CDK Global, 605 Fifth Ave S, Ste 800, Seattle, WA 98104 e-mail: jason.kessler@gmail.com Nicolas Nicolov J.D. Power and Associates / McGraw-Hill, 4888 Pearl East Circle, Boulder, CO 80301 e-mail: nicolas nicolov@yahoo.com Work was conducted while both authors were at the Web Intelligence Division, J.D. Power and Associates / McGraw Hill. 1