Gregory Todd Williams SPARQL BGP Optimization Presentation Shangguan 2

From Semantic Portal Wiki

• Question is for the Presentation: Gregory Todd Williams SPARQL BGP Optimization Presentation
• Question is asked by: Zhenning Shangguan
• The Question is: In section 5.1, the author argues that "the selectivity of a triple pattern is estimated by the formula sel(t)=sel(s)*sel(p)*sel(o)...", and later he states that "...Note that this formulationonly approximates sel(t) as it implicitly assumes that sel(s), sel(p), and sel(o) are statistically independent, which they will not be in most cases." What's the point of saying this? If in practical the assumption of statistical independence between subject, predicate, and object does not always hold, does it mean that this fomula is of little use? And also, what about another formula of selectivity estimation which looks like this: sel(t)=c1*sel(s)+c2*sel(p)+c3*sel(o), where c1-c3 help to normalize sel(t) to 0-1?

In section 5.1, the author argues that "the selectivity of a triple pattern is estimated by the formula sel(t)=sel(s)*sel(p)*sel(o)...", and later he states that "...Note that this formulationonly approximates sel(t) as it implicitly assumes that sel(s), sel(p), and sel(o) are statistically independent, which they will not be in most cases." What's the point of saying this? If in practical the assumption of statistical independence between subject, predicate, and object does not always hold, does it mean that this fomula is of little use? And also, what about another formula of selectivity estimation which looks like this: sel(t)=c1*sel(s)+c2*sel(p)+c3*sel(o), where c1-c3 help to normalize sel(t) to [0-1]?