An Information Retrieval Model Based On Word Concept

An Information Retrieval Model Based On Word Concept Chen Wu1，2, Quan Zhang2, Xiangfeng Wei2 1 Graduate School of Chinese Academy of Sciences, 100039, Beijing, China 2 Institute of Acoustics, Chinese Academy of Sciences, 100080, Beijing, China wuchen@mail.ioa.ac.cn Abstract. Traditional approaches for information retrieval from texts depend on the term frequency. A shortcoming of these schemes, which consider only occurrences of the terms in a document, is that they have some limitations on extracting semantically exact indexes that represent the semantic content of a document. However, one word can always represent more than one meaning. The word sense ambiguities will also affect the system behavior. To address this issue, we proposed a brand new strategy – a concept extracting strategy to extract the concept of the word and to determine the semantic importance of the concepts in the sentences via analyzing the conceptual structures of the sentences. In this approach, a conceptual vector space model using auto-threshold detection is proposed to process the concepts, and a cluster searching model is also designed. This auto- threshold detection method can help the model to obtain the optimal settings of retrieval parameters automatically. An experiment on the TREC6 collection shows that the proposed method outperforms the other two information retrieval (IR) methods based on term frequency (TF), especially for the lower-ranked documents Keywords: Information Retrieval, Word Concept, Sentence Category, Auto-threshold detection clustering method 1 Introduction Searching in archives of documents becomes increasingly frequent for most of people. So, how to provide useful and efficient IR (Information Retrieval) tools becomes more and more important. Since 1968, when the first formal model for IR[1] comes into being, a number of IR models have been developed, such as vector space, probabilistic, fuzzy, logical, inference, language and so on[2-10].While considering all the former methods, we can find most of the methods are based on occurrences of terms in a document (or TF: Term Frequency) and seldom on the content of the document. These algorithms analyze only term occurrences and do not attempt to resolve the meaning of the terms. As we all know, the meaning of words is very helpful to IR. If we disregard the context and dispose the words separately, the performance of IR system will be lowered for the ubiquitous existence of word sense ambiguity. A word may have a lot of meanings. A particular meaning can also be represented by more than one word. Therefore whether we can determine the meaning of the word in a document will affect the accuracy of an IR system. Meanwhile, many studies[11-15] have shown that people understand things by comprehending the concepts represented by the things. The language works in the same way[16]. Consequently, some research teams began to investigate the language conceptual space and the expression of the space using symbolic system. The main part of the IR method we present is to discuss how to draw the conceptual expressions of a word and a sentence based on a symbolic system[16-19]. We consider the advantages of case …

Content truncated to 500 words