A practical approach to Boosting Documents with Solr/Lucene

By definition, document boosting is anything but a “one size fits all” phenomenon, observes Tim Potter of National Renewable Energy Laboratory (NREL), in his talk on “Boosting Documents in Solr by Recency, Popularity, and Personal Preferences” on Day 1 of Lucene Revolution.

Slides for this session.

Tim’s talk outlined a number of innovative and improved approaches to boosting documents in Solr by age and popularity. He also covered document filtering based on user preferences. For each of his topics, he presented Tips and Tricks, and a constant excellence of his presentation was the sample code snippets heavily sprinkled throughout.

Tim noted that boosting is about raising the relevance score, and the amount by which it is raised must be appropriate to the metric used. His approach to boosting by age involves the use of the recip and ms query functions, and he showed some graphs that compared results obtained by using different scoring adjustments.

The next method, boosting by popularity, can directly compete with boosting by age. Generally, the older a document gets, the less boosting it will receive. The “competition” typically happens when a document is pretty old but experiences a spike in subsequent popularity for whatever reason. Thus, as Tim emphasized, it is important when boosting by popularity to look at that popularity in the context of “time slots.”

Tim’s final comments on filtering by user preferences came with the caveat that the most important thing to avoid when implementing this technique is over-exclusion, which can easily be a consequence of this technique. His approach depends on the use of a Solr FastLRUCache, and requires a determination of whether the user’s preferences have changed, requiring a cache flush.

Cross-posted with Lucene Revolution Blog. Tony Barreca is a guest blogger.This is one of a series of presentation summaries from the conference.