Minimally Supervised Learning of Semantic Knowledge from Query Logs

Minimally Supervised Learning of Semantic Knowledge from Query Logs Mamoru Komachi Hisami Suzuki Nara Institute of Science and Technology Microsoft Research 8916-5 Takayama One Microsoft Way Ikoma, Nara 630-0192, Japan Redmond, WA 98052 USA mamoru-k@is.naist.jp hisamis@microsoft.com Abstract We propose a method for learning semantic categories of words with minimal supervi- sion from web search query logs. Our me- thod is based on the Espresso algorithm (Pantel and Pennacchiotti, 2006) for ex- tracting binary lexical relations, but makes important modifications to handle query log data for the task of acquiring semantic categories. We present experimental results comparing our method with two state-of- the-art minimally supervised lexical know- ledge extraction systems using Japanese query log data, and show that our method achieves higher precision than the pre- viously proposed methods. We also show that the proposed method offers an addi- tional advantage for knowledge acquisition in an Asian language for which word seg- mentation is an issue, as the method utiliz- es no prior knowledge of word segmenta- tion, and is able to harvest new terms with correct word segmentation. 1 Introduction Extraction of lexical knowledge from a large col- lection of text data with minimal supervision has become an active area of research in recent years. Automatic extraction of relations by exploiting recurring patterns in text was pioneered by Hearst (1992), who describes a bootstrapping procedure for extracting words in the hyponym (is-a) relation, starting with three manually given lexico-syntactic patterns. This idea of learning with a minimally supervised bootstrapping method using surface text patterns was subsequently adopted for many tasks, including relation extraction (e.g., Brin, 1998; Ri- loff and Jones, 1999; Pantel and Pennacchiotti, 2006) and named entity recognition (e.g., Collins and Singer, 1999; Etzioni et al., 2005). In this paper, we describe a method of learning semantic categories of words using a large collec- tion of Japanese search query logs. Our method is based on the Espresso algorithm (Pantel and Pen- nacchiotti, 2006) for extracting binary lexical rela- tions, adapting it to work well on learning unary relations from query logs. The use of query data as a source of knowledge extraction offers some unique advantages over using regular text.  Web search queries capture the interest of search users directly, while the distribution of the Web documents do not necessarily reflect the distri- bution of what people search (Silverstein et al., 1998). The word categories acquired from query logs are thus expected to be more useful for the tasks related to search.  Though user-generated queries are often very short, the words that appear in queries are gen- erally highly relevant for the purpose of word classification.  Many search queries consist of keywords, which means that the queries include word segmenta- tion specified by users. This is a great source of knowledge for learning word boundaries for those languages whose regularly written text does not indicate word boundaries, such as Chi- nese and Japanese. Although our work naturally fits into the larger goal of building knowledge bases …

Content truncated to 500 words