MIMS Final Project 2008

cocobean: A Socially Ranked Question Answering System

Advisor: 

Search in its current form is still in its infancy. Finding exactly what one wants is usually a process of constantly refining ones query with additional words. Traditional search engines have no inherent understanding of what a user wants but matches documents to queries based on presence or lack of presence of search terms. In natural language search, a user prompts the system with a coherent question and the system returns a single or a set of candidate answer sentences. By making search more natural and intuitive, natural language search is fundamentally changing how we search the web and delivering higher quality results.

Our project consists of a natural language search engine that utilizes both algorithmic search as well as methods for users to socially influence the ranking of results. The main components of our system consist of a traditional question answering system based on a paper by Microsoft Research. The paper calls for the creation of question-to-answer transformations, essentially to figure out how a candidate answer might look like given a question. We will be using the Lucene search engine to find relevant documents in our corpus and to extract candidate answers. We then plan to learn which transformations worked well by pairing candidate answers with their transformation and having users vote on the ‘correctness’ of the answer. The results will then be reordered based on both algorithmic rank and social rank.

Parts of our project was developed for Information Retrieval and Search Engines and Social Media. We plan to expand on our existing work of using manually created transformation with automatically learned ones. In addition, we wish to utilize more natural language processing techniques such as entity detection and co-reference resolution to build a semantic index.

More Information

Last updated:

October 7, 2016