Improve Primo Search Algorithms
When our users look for a book in the Primo Search Portal they expect that the Primo Search behaves like a Google Search, i.e. they expect that Primo finds books with titles or metadata containing words similar to those in their query. Unfortunately, the Primo Search is very sensitive to the exact spelling of the words in the query. If the search query is slightly inaccurate, the book is not found in Primo. I tried such unsuccessful search queries on www.amazon.de, and Amazon usually found the correct book: Amazon showed the correct hit on top of the list (see attachment). This shows that the search algorithms of Amazon are clearly superior to those of Primo. I feel that Ex Libris should take steps to keep up with modern search technologies.
Some, but not all, of Primo’s shortcomings seem rather easy to fix. Primo fails to take word flexions in non-English languages into account. Since the endings of German words not only depend on plural and singular but also on the context, Primo’s search results for German queries are overly sensitive to word flexions. However, some of the English examples indicate that improving Primo’s dictionaries will not be sufficient. Over the long term, ExLibris may consider to team up with a specialized search provider.
Similar proposals in ExLibris idea exchange
https://ideas.exlibrisgroup.com/forums/308176-primo/suggestions/37586617-increase-primo-search-capability-to-support-major
https://ideas.exlibrisgroup.com/forums/564566-summon/suggestions/17415091-automatically-searching-for-the-corrected-spelling
Roland Suri, ETH-Bibliothek
shared this idea
-
Sarah
commented
This idea was shared in 2019 and has gotten a fair amount of support. Is there any update?
-
Roland Suri, ETH-Bibliothek
commented
The search results of Primo VE turned out to be worse than those of the Ex Libris Summon Search. Indeed, according to the product documentation of Ex Libris the search algorithms of Summon are more elaborate than those of Primo: Only with Primo VE (bu tnot with Summon) stemming is limited to queries that produce less than 25 hits. Only Summon (but not Primo VE) provides tokenization and decompounding. Only Summon provides "Verbatime Match Boost". I feel that these more elaborte search methods of Summon lead to better search results. I thus hope that Ex Libris will improve the Primo VE search methods using the Summon search methods.
-
Roland Suri, ETH-Bibliothek
commented
I just tested the retrieval performance of several search engines for English books using difficult search queries. The queries contained spelling variants, typos, word inflections and synonyms. Google Books came out best (100% hits). The Amazon Book Search (53 %), the EBay Book Search (43 %) and the Ex Libris Summon Search (40 %) scored lower. Primo VE performed badly (21 %) (see both attached documents for details).
Unfortunately, we cannot use Google to search our collections. The EBay Book Search uses the open source software Elastic Search, but the configuration of this software for the ETH library would be very demanding. The Ex Libris Summon Search would be an acceptable solution.