APPLICATION OF FOURIER TRANSFORM TECHNIQUES TO IMPROVE INFORMATION RETRIEVAL IN DIGITAL LIBRARIES

The growing volume of digital information calls for better approaches to extracting relevant documents in digital libraries. This research explored the effectiveness of Fourier transform techniques in improving retrieval via frequency domain filtering of TF-IDF vectors of high dimensionality. The baseline MAP@10 was obtained using the Cranfield Collection dataset (consisting of 1,400 documents, 225 queries, and 1,837 relevance judgments). This study evaluated Fourier Transform filtering of TF-IDF representations for information retrieval. The baseline MAP@10 was 0.2232. Fast Fourier Transform filter was applied at six cutoffs ranging from 0.05 to 0.50. As performance metrics, we used Precision@10, Recall@10, and MAP@10. Statistical significance was calculated using the Wilcoxon signed-rank test with Holm correction and bootstrap confidence intervals. The highest MAP@10 value (0.2403, an increase of 7.64%) was recorded for the 0.20 cutoff, but it did not reach statistical significance (adjusted p = 0.0934). There were significant increases in MAP@10 at cutoff values of 0.30 (0.2383, adjusted p = 0.0269) and 0.40 (0.2375, adjusted p = 0.0026). These results indicate that frequency filtering may help improve retrieval in certain cases. This approach should be further examined in Nigerian academic libraries in view of OCR errors and insufficient infrastructure. Possible connections with Nigeria’s NDPA 2023, NITDA AI Strategy 2024, and SDGs 4 and 9 are associated with the study.