Introduction

Most of the problems in information retrieval systems occur from three sources: impreciseness of query and document representations, the changing state of mind in document relevant judgement, and the discrepancy of retrieval technique to match query and information need. The traditional approach to information retrieval systems, such as Boolean based retrieval technique, cannot solve these problems (Belkin and Croft, 1987). According to McCain (1989) and Pao and Worthen (1989), retrieval by keywords and cited documents will end up with different sets of documents retrieved. There were relevant documents retrieved by keywords but not retrieved by references, and vice versa. Hence, a new approach to retrieval technique is needed. This study is intended to build a new approach to information retrieval systems by employing the inherent structure of a document collection in an effort to learn more about document components that might improve information retrieval performance. The document components examined are the pattern of keywords, citing documents, and cited documents. Three independent variables were studied: co-keyword, co-citing document, and co-cited document (Hasibuan, 1995). These three variables constitute the multi-dimensions concept-based information retrieval system. By providing such variables as entry points to search relevant documents, it is widening the naturalness of the system in order to accommodate users' information needs.

Methodology

A test collection was constructed from a collection of research articles published by National Atomic Research Agency (BATAN), covering the period 1985 through 1998. An automatic program was written to build indexes of keywords, citing documents, and cited documents for each document. The relationship that may occur between two documents can be depicted as in Figure 1. Pair-wise document similarity is calculated on those three variables. As in Figure 1, the similarities between documents A and B can be viewed in terms of documents Q and S, which cite documents A and B (co-cited), and documents Y and Z, which are cited by A and B (co-citing). In addition to that, the similarity of documents A and B can be counted on the number of shared terms (co-keyword).

Figure 1. Figure 1. Relationship of Two Documents (A and B)

Document similarity is measured by using the simple matching coefficient proposed by Van Rijsbergen (1979) and Salton (1989). The similarity of document A and B is calculated as follows:
Similarity (A, B) = | X Ç Y | The variables X and Y represent the sets of index terms occurring in two documents. Hence, the similarity of document A and B is measured in terms of the number of shared index terms. These sets of index terms can be expanded to include values for citing documents and cited documents. The retrieval technique used is based on this document similarity.

Results

The preliminary results of the research showed that relevant documents retrieved by one component did not always agree with other components (see Figure 1). Figure 1 shows document 0376 and document 0419 have 11 shared index terms, two shared cited documents and three shared citing documents. As we expected, most of the document pairs have zero frequency of shared cited document and shared citing document. This finding is in line with our previous research finding in Hasibuan (1995). According to these results, it is suggested to build a multi-dimension concept-based retrieval system. The system built is able to provide users with a facility to search several interrelated options of search strategy. Given that kind of facility, a user can be more flexible to start his/her search by using one of the dimensions, says keyword, then navigates the system using other dimensions of document similarity. At any moment, the documents retrieved can be viewed, evaluated and judged for their relevance.

Figure 2. Figure 1. A Portion of Search Results of the Multi-dimension Concept-based Information Retrieval System

The search results shown in Figure 1 are posted in a hypertext based, so that a user can continue browsing uninterruptedly, in order to further his/her search to retrieve more on the other possible relevant documents. For each pair of documents retrieved, the system will provide the frequency of its co-keyword, co-cited documents and co-citing documents. Furthermore, the non-zero frequency of each entry in the columns of co-citing and co-cited documents becomes an active icon. If we click 2 in column co-cited, then we can see the documents that are co-cited by document 0376 and 0419 (see Figure 2). There are documents 0523 and 0531. The abstract of each document retrieved can be viewed by clicking the number of the document (see Figure 3).

Figure 3. Figure 2. An Example of Co-cited Documents

Figure 4. Figure 3. An Example of Abstract Document Retrieved

Conclusion

Multi-dimensions concept-based retrieval system provides relaxed facility in order to search information by utilizing the components of documents - co-keyword, co-cited documents, and co-citing documents. With this facility, the system can widely accommodate the range of user's search strategy in information seeking. The drawback of the system compared to the traditional system is that this new approach needs more space of computer storage in order to accommodate all its index files. Furthermore, it can slow down the search process. However, this trade-off is compensated for by the flexibility of the system to provide more search strategies, more comprehensive retrieval of relevant documents, and easy to browse from one document to another document. Ultimately, by utilizing these three components of a document, the system can reduce the possibility of low retrieval system performance due to the impreciseness of query and document representations, lack of relevant judgement, and lack of matching function between query and document.

References

Nicholas J.Belkin W. BruceCroft. “Retrieval Techniques.” Annual Review of Information Science And Technology. 1987. 22: 109-145.

ZainalA.Hasibuan. “Document Similarity and Structure: Using Bibliometric Methods and Index Terms As Approaches to Improving Information Retrieval Performance.” School of Library, Archive and Information Science, Indiana University, Bloomington, Indiana, 1995.

Katherine W.McCain. “Descriptor and Citation Retrieval in the Medical Behavioral Sciences Literature: Retrieval Overlaps and Novelty.” Journal of the American Society for Information Science. 1989. 40: 110-114.

Miranda L.Pao Dennis B.Worthen. “Retrieval Effectiveness by Semantic and Citation Searching.” Journal of the American Society for Information Science. 1989. 40: 226-235.

Gerard Salton. Automatic Text Processing: The Transformation, Analysis and Retrieval of Information by Computer. New York: Addison-Wesley Publishing Company, 1989.

S. R. Van Rijsbergen. Information Retrieval. London: Butterworth, 1979.