Hiroko Mano

Selecting expansion terms in automatic query expansion

In automatic query expansion, where queries are automatically expanded with terms not in the original query but extracted from initially retrieved top-ranked documents, each term in the top-ranked documents is evaluated for its usefulness as an expansion term to be added to the original query. As the evaluation measure, we have used since [1] a Term Selection Value function based on a combination of a relevance weight [3] derived from document frequencies and a sum of within-document term frequencies, so that terms that are speci c to top-ranked documents (relevance weight factor) and representative of each top-ranked document (term frequency factor) would be rated highly. While query expansion based on the TSV function above contributed to improvement in retrieval e ectiveness in the past experiments, recent experiments suggested selecting expansion terms based on both the relevance weight and the term frequencies might not always be the best strategy; speci cally, when the top-ranked documents, assumed to be relevant to the topic, turn out to be mostly non-relevant, the relevance weight seemed to lead to inappropriate selection of expansion terms. In this paper, we investigate how a relevance weight affects expansion term selection as the number of relevant documents in the top-ranked documents decreases and examine the e ectiveness of an alternative approach of not using a relevance weight in expansion term evaluation.

TRMeister: a DBMS with high-performance full-text search functions

TRMeister is a DBMS with high-performance full-text search functions. With TRMeister, high-speed full-text search, including high-precision ranking search in addition to Boolean search, is possible. Further, in addition to search, high-speed insert and delete are possible, allowing full-text search to be used in the same way as other types of database search in which data can be searched right after data is inserted. This makes it easy to combine normal attribute search with full-text search and thus easily create text search applications.

Ricoh at CLEF 2003

This paper describes RICOH’s participation in the Monolingual Information Retrieval tasks of the Cross-Language Evaluation Forum (CLEF) 2003. We applied our system using the same kind of stemmer, the same options and different parameters to five European languages and compared the results for each langauge. Although the overall performance of the system was reasonable, there were two problems. The first was the lack of a compound splitter for German and the second was the failure of query expansion when there were few relevant documents.

The ECHO method: concurrency control method for a large-scale distributed database

When constructing a large-scale database system, data replication is useful to give users efficient access to that data. However for databases with a very large number of replica sites, using traditional networking concurrency control protocols makes communication too costly for practical use. In this paper we propose a new protocol called ECHO for concurrency control of replica data distributed on a large scale. Using broadcasting, such as satellite broadcasting, terrestrial broadcasting, and cable television, the ECHO method can keep communication costs constant, regardless of the number of sites. The ECHO method also has a backup mechanism for failed broadcast reception using the communication network.