This appendix describes seven search parameters that can affect search optimization. For most applications, these settings have little effect, because the search space is small to begin with. However, in very large applications, these settings greatly help system performance with little or no degradation in search quality.
estimateDocMaxlimits the total number of index items to consider as possible candidates. This setting can prevent the explosion of candidates caused by very common terms. Since ATG Search sorts the query terms by inverse frequency, ATG Search will use the most-specific (high-content) terms first, and if the candidates exceeds this setting, it will usually be with the very common (low-content) terms, without any loss of relevant results.maxDocumentslimits the number of documents to inspect for relevant statements. Since ATG Search sorts the document candidates by relevancy, this setting will exclude the least relevant documents, most likely with little loss of relevant results. However, it is possible that a low-relevance document has a very relevant statement which would be lost.estimateMaxlimits the number of statement candidates to collect during search. This is analogous toestimateDocMax, and strives to prevent the explosion of statement candidates caused by very common terms. Again, since ATG Search sorts the query terms by inverse frequency, ATG Search will use the most-specific (high-content) terms first, and if the candidates exceeds this setting, it will usually be with the very common (low-content) terms, without any loss of relevant results.retMaxlimits the number of statement candidates to perform vector matching on. The vector matching is computationally expensive, so limiting the matching improves speed. The statement candidates are sorted by an estimation metric, so this setting limits the candidates to the most relevant.retLimitis a cut-off on the number of relevant statement matches to collect. The vector matching is computationally expensive, so limiting the matching improves speed. The statement candidates are sorted by an estimation metric, so this setting will limit the candidates to the most relevant.The maximum result pool, specified by the
poolparameter, limits the number of matching statements that will be collected for final result processing. Since typically applications show a limited number of results, this setting represents a reasonable estimate of how many raw results it will take to construct the final results, with respect to grouping and duplicate handling. This setting prevents the wasted processing to construct thousands of results, just to return the first page of 10.The maximum results by type, specified by the
typeparameters, controls the final grouping, sorting, and duplicate handling. These settings do not really reduce the search space, although they can reduce the number of results to process at this final stage.
