POSTECH at NTCIR-5 Patent Retrieval: Smoothing Experiments in a Language Modeling Approach to Patent Retrieval

This report describes the experimental results of our participation at the Document Retrieval Subtask of NTCIR-5 Patent Retrieval Task. Unlike newspaper articles which belong to the main document type handled in previous information retrieval experiments, patent documents have many different characteristics in terms of length, technicality, structureness, etc. Among these, we focus on the length of patent documents. Since patent documents are long and cover a diverse spectrum of topicality, document models estimated from them by the language modeling approach are expected to show different characteristics from those estimated from short document samples such as newspaper articles. Based on such contemplation, this report investigates the effect of smoothing in the language modeling approach on retrieving patent documents.