PatentCom: A Comparative View of Patent Document Retrieval

Abstract

Patent document retrieval, as a recall-orientated search task,

Comparative Document Analysis for Large Text Corpora

“…1, a citation recommendation system can show users the common and distinct concepts produced by CDA to help them understand the connections and differences between a query paper [11] and a recommended paper [6]. In a similar way, ... CDA can reduce the efforts on patentbility searching [31].…”

“…Comparative document summarization [31,19,10,29,34,9] focuses on the distinction aspect-it summarizes the differences between comparable document sets by extracting the discriminative sentences from each set. Existing work formalizes the problem of discriminative sentence selection into different forms, including integer linear programming [10], multivariate normal model estimation [29], and group-related centrality estimation [20].…”

“…As shown in our experiments, this method (labeled as Word-Match) suffers from low recall in finding common terms and low precision in finding distinct terms since it ignores semantic common phrases and pairwise distinct phrases. Similarly, Zhang et al [31] consider both common and distinct aspects in generating comparative summary for patent documents. They derive a term co-occurrence tree which can be used to extract sentences for summarization.…”

“…6). Recent work [31] considers pairwise distinc-tion property by casting the problem as feature selection but it appears to produce many overly specific phrases.…”

“…Instead of measuring the similarity between items or predicting sentiment rating on an item, it delves deeper into the text and focuses on extracting content units of different granularity (e.g., words, sentences) to summarize the commonalities and differences. With the rapid emergence of text-based data in many domains, automatic techniques for comparative text mining have a wide range of applications including social media analysis (e.g., opinion mining on twitter [27,18]), business intelligence (e.g., customer review analysis [14,12], news summarization [10,25,16]), and scientific literature study (e.g., patentability search [31] ... [11] Scaling Personalized Web Search [6] Topic-sensitive PageRank topic-sensitive@@ranking vector; context-specific importance score; topic-sensitive PageRank; query context; ... Figure 1: Example output of CDA for papers [11] and [6].…”