SBIR Phase I: Semantic Information Extraction From Text

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project can be summarized as follows. (1) Businesses will benefit from the proposed product because companies are increasingly using data-driven predictive models to improve their bottom line. In many areas where structured data are readily available, such as credit scoring, these models have produced impressive returns on investment. One of the main obstacles hindering the application of these models in other areas is the lack of structured data. By extracting information from sources such as web pages, blogs and social media messages, and storing it as structured data, companies will be able to take advantage of the vast amount of unstructured data that are generated daily and thereby improve their bottom line. (2) As the Chinese economy expands and becomes more deeply intertwined with the US economy, many US businesses will need high-quality and timely information about Chinese markets. However, information extraction from Chinese text is an underserved area. The proposed product can fill this void and thereby meet a significant commercial demand. This Small Business Innovation Research (SBIR) Phase I project will develop a new information extraction (IE) method. Currently IE focuses on information extraction from short text snippets consisting of a few words in order to derive structured factual information from unstructured text. But its performance is often deteriorated by the shortage of features - the sparse feature problem. A major benefit of the approach developed in this project is that it takes advantage of the important role of specific linguistic units even when the number of words in a sentence is limited. These linguistic units give a sentence its structure. Depending on structural characteristics or functional principles of sentences, the features around them are grouped. The grouped features satisfy certain properties and can be used to capture structural and semantic information, which is helpful for minimizing the sparse feature problem. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.