Combining Positive and Negative Query Feedback in Passage Retrieval

Information Retrieval Systems aim at retrieving relevant documents according to the information needs which users express. Most Information Retrieval Systems focus on passage retrieval where the granularity of information retrieved is not the document but a smaller unit such as a sentence or passage. These systems try to better answer the users' needs by giving more importance to the most relevant document parts. This paper addresses the problem of passage retrieval as defined by the TREC novelty track, subtask 1 where the aim is retrieving relevant sentences from relevant documents. We define a new term weighting function that takes non relevancy information into account and which is based on query evidence only meaning that it does not need global parameters such as tf.idf term weights. Our method is evaluated on both the 2002 and 2003 TREC novelty collection where we show that taking into account the narrative part that describes nonrelevant documents is useful as well as is e...

• Type: Conference
• Year: 2004
• Where: RIAO
• Authors: Taoufiq Dkaki, Josiane Mothe
• Keywords: Information Retrieval Systems, Passage Retrieval, Relevant Documents, RIAO 2004, RIAO 2007