What does the SEARCH_RELEVANCE_FOCUS configuration setting do?
Environment:
Oracle B2C Service all versions
Resolution:
SEARCH_RELEVANCE_FOCUS boosts the weight of clustered answers and specifies the degree of conceptual matching between the search query and results using the learned topics from within the knowledge base. A setting of 0 applies only literal search matching. Settings from 1 to 3 apply increasingly more bias toward matching learned conceptual topics in the search results. Default is 0.
There are three main parts that affect this configuration setting:
#1. The search terms entered by the end user
#2. The search results, which are answer IDs that matched the search terms. Each result has a certain 'match weight'
#3. The 'learned topics', which is another term for answer clusters. Answer clusters are groups of answers that share conceptual similarity, with representative words stored in the cluster_class table.
How it works:
If a search result (#2) is in a cluster (#3) that contains one of the search terms (#1), then the result's match weight is boosted by how closely the result is associated with that cluster. The SEARCH_RELEVANCE_FOCUS configuration setting adds additional boosting (to the weight) to that match.
In other words, when you search on some words, and those words are contained in answer clusters, then the answers in those clusters get a boost in the search rankings.
Things to keep in mind:
- If this configuration setting is zero, this entire process is skipped.
- The 'learned topics' are only enabled if the CLUST_ENABLED setting is true.
- The particulars of how clusters are constructed, and exactly how much the setting affects the boosting are part of the proprietary Knowledge Foundation (KF) algorithm. We can say that the boosting, due to this cluster matching, can be "significant" depending on how similar the search result is compared to the answer cluster (if enabled). On the other hand, the boost difference between the configuration's three values is modest: a linear amount.
