Tuomo Korenius’s research while affiliated with Tampere University and other places

Clustering groups document objects represented as vectors. An extensive vector space may cause obstacles to applying these methods. Therefore, the vector space was reduced with principal component analysis (PCA). The conventional cosine measure is not the only choice with PCA, which involves the mean-correction of data. Since mean-correction changes the location of the origin, the angles between the document vectors also change. To avoid this, we used a connection between the cosine measure and the Euclidean distance in association with PCA, and grounded searching on the latter. We applied the single and complete linkage and Ward clustering to Finnish documents utilizing their relevance assessment as a new feature. After the normalization of the data PCA was run and relevant documents were clustered.

Search facilitated with agglomerative hierarchical clustering methods was studied in a collection of Finnish newspaper articles (N = 53,893). To allow quick experiments, clus- tering was applied to a sample (N = 5,000) that was reduced with principal components analysis. The dendrograms were heuristically cut to find an optimal partition, whose clusters were compared with each of the 30 queries to retrieve the best-matching cluster. The four- level relevance assessment was collapsed into a binary one by (A) considering all the relevant and (B) only the highly relevant documents relevant, respectively. Single linkage (SL) was the worst method. It created many tiny clusters, and, consequently, searches enabled with it had high precision and low recall. The complete linkage (CL), average linkage (AL), and Ward's methods (WM) returned reasonably-sized clusters typically of 18 -32 doc uments. Their recall (A: 27-52%, B: 50-82%) and precision (A: 83-90%, B: 18-21%) was higher than and comparable to those of the SL clusters, respectively. The AL and WM clustering had 1- 8% better effectiveness than nearest neighbor searching (NN), and SL and CL were 1-9% less efficient that NN. However, the differences were statistically insignificant. When evaluated with the liberal assessment A, the results suggest that the AL and WM clustering offer better retrieval ability than NN. Assessment B renders the AL and WM clustering better than NN, when recall is considered more important than precision. The results imply that collections in the highly inflectional and agglutinative languages, such as Finnish, may be clustered as the collections in English, provided that documents are appropriately preprocessed.

Stemming and lemmatization were compared in the clustering of Finnish text documents. Since Finnish is a highly inflectional and agglutinative language, we hypothesized that lemmatization, involving splitting of the compound words, would be more appropriate normalization approach than the straightforward stemming. The relevance of the documents were evaluated with a four-point relevance assessment scale, which was collapsed into binary one by considering all the relevant and only the highly relevant documents relevant, respectively. Experiments with four hierarchical clustering methods supported the hypothesis. The stringent relevance scale showed that lemmatization allowed the single and complete linkage methods to recover especially the highly relevant documents better than stemming. In comparison with stemming, lemmatization together with the average linkage and Ward's methods produced higher precision. We conclude that lemmatization is a better word normalization method than stemming, when Finnish text documents are clustered for information retrieval.