Comparison of Cosine Similarity and BM25 in Book Procurement Prioritization Based on Multi-Source Data Integration

Book collection is one of the important components in a library. Providing book collections is a challenge for libraries to ensure that these collections are in accordance with the needs of the user and the development of knowledge. To accomplish this, multi-source data integration is used, consisting of borrowing data, user suggestion data, and trend data. The results of the integration of multiple data sources in the form of keywords are implemented using the cosine similarity and BM25 methods to provide a list of book recommendations that are used as procurement priorities. Cosine similarity provides recommendations for 728 book titles and BM25 provides recommendations for 82 book titles. BM25 precision results get a value greater than cosine similarity which is 0.965, while the comparison of recall results cosine similarity is greater than BM25 which is 0.965. From the recommendation and comparison results, it can be concluded that cosine similarity provides a larger scope of book title recommendations than BM25 recommendations which are more selective. The larger scope of cosine similarity book title recommendation results allows the library to hold a more diverse collection and is appropriate for large libraries. While the more selective BM25 recommendation results can help with the procurement of focused collections, they are best suited for libraries with special collections.