An Effective Multilingual Document Retrieval System for Electronic Business

[[abstract]]隨著這幾年網際網路的普及與發展，從網路上我們可以得到各式各樣的資訊，面對這些繁雜又龐大的資訊。使用者要如何找到最符合需求的資訊並且不必經過冗長的資料收集與尋找，跨語言資訊檢索（Cross-Language Information Retrieval CLIR）就是為了此目的而生。使用者經由輸入關鍵字或是語句就能擷取到最適當的資訊，多語言資訊檢索（Multi-Language Information Retrieval MLIR）其實就是跨語言資訊檢索的進階搜尋。以往跨語言資訊檢索提供一對一的檢索模式而多語言資訊檢索提供的是一對多的檢索模式，其差異在於多語言資訊檢索模式在語言翻譯時原始語言（Source Language）必須藉由中繼語言（Meta Language）的轉換來查詢目標語言（Target Language）。但是此兩種檢索模式最主要的目的不外乎是幫助使用者使用某一種特定語言做查詢，同時可以檢索出不同語言的相關文件，使用者不需要自行翻譯搜尋的字或語句，使用自己熟悉的語言便能夠檢索出其他語言的文件。 在本論文中，我們顛覆傳統不再只是針對使用者下的Query進行分析與查詢擴充，我們將重點著重在文件的分析與索引擴展（Index Expansion）上。使用句構分析模型（Sentence Structure Analysis Model SSAM）來取代傳統的斷詞方法，更準確的擷取出能夠表現文件內容的索引，再來使用知識本體論模型(Ontological-Model)來體現文件與文件之間語義網路的關係。 我們使用NTCIR第四屆與第六屆所提供之查詢集，包含繁體中文、英文、日文等文件集來作為測試資料。這些文件經過我們提出之研究模型分析後，實驗結果顯示，我們提出之句構分析模型能夠比之前的研究提升平均檢索準確度39%，單語資訊檢索可以比之前的研究提升了47%。[[abstract]]The main purpose of Cross-Language Information Retrieval (CLIR) is to help users find the information to meet their need By using the proposed system users enter keywords or a statement would be able to retrieve the most appropriate information Multi-Language Information Retrieval (MLIR) is actually the advanced search of CLIR In the past CLIR provided one language to another language’s information retrieval model Yet multilingual information retrieval models provided users to use one language to retrieve multiple languages The main difference of CLIR and MLIR is that multi-language information retrieval model need to translate Source Language to Meta Language and then use Meta Language to retrieve Target Language However the goal of these two kinds of retrieval models is to help users to obtain the information from specific query language and do not need to translate the query terms to other languages In this paper without dealing with the Query Terms and Query Expansion we focus on Document indexing and proposed the concept of Index Expansion Word segmentation is replaced by Sentence Structure Analysis Model Using the Ontological-Model to reflect the relationship between document and document obtain indexes of documents more accurately In our research we use the data sets provided by the 8th NII Test Collection for IR Systems (NTCIR) including the languages of Traditional Chinese Simplified Chinese and Japanese as the test data set The results show that our proposed method effectively enhances the retrieval accuracy of multi-language document retrieva