
We are excited to announce the fifth issue dedicated to the Principles of Database Systems (PODS) research track of the Proceedings of the ACM on Management of Data (PACMMOD) journal. In its current form, the journal hosts a SIGMOD and a PODS research track. The PODS research track aims to provide a solid scientific basis for methods, techniques, and solutions for the data management challenges that continually arise in our data-driven society. Articles for the PODS track of PACMMOD present principled contributions to modeling, application, system building, and both theoretical and experimental validation in the context of data management. Such articles might be based, among others, on establishing theoretical results, developing new concepts and frameworks that deserve further exploration, providing experimental work that sheds light on the scientific foundations of the discipline, or conducting a rigorous analysis of industry artifacts that are widely used and/or recently developed. At a time when computer science is increasingly data-centric, it is essential to promote an active exchange of tools and techniques between principles of database systems and other communities focused on data management. The PODS research track thus pays special attention to articles that promote the integration of data management techniques within broader computer science.
Enumerating all matches of a regular expression with capture variables (variable regex, for short) to a document is a fundamental task in textual information extraction, e.g., in connection to document spanners. State-of-the-art results [Amarilli et al., ACM TODS 2021] show how the matches of a variable regex, representable by a sequential variable-set automaton, to a document can be enumerated with delay constant in the size of the document and polynomial in the size of the query, after a preprocessing requiring time linear in the size of the document and polynomial in the size of the query. However, the polynomial dependency on the size of the query in both the delay and the preprocessing is actually quite high: at least quadratic. We consider here the matching problem for regular expressions with capture variables which can be represented as regular patterns: w 0 x 0 w 1 x 1 #8230;s w k x k w k +1, where, for i ∈ {0, … k +1}, w i is a string of terminal letters and, for i in {0,…, k }, x i is a variable. This problem naturally lies in the intersection of information extraction and stringology, as it can also be seen as computing all the ways in which k +1 given strings w 0 ,…,w k +1 occur, in this order and without overlaps, in a given text T . We provide an optimal enumeration algorithm for this problem. After a preprocessing requiring linear time in the size of the document T and the size of the query pattern α, we can enumerate all matches with worst-case constant delay (in combined complexity, i.e., both in the document- and the query-size).
Query optimizers must efficiently choose a query plan using noisy estimates of cardinalities. We study robust query optimization, where the optimizer is made aware of the uncertainty in cardinality estimates and needs to select the most ''robust'' plan. A key empirical observation is that a small set of plans often suffices as robust plan candidates for all queries following a given template. We formalize this phenomenon by introducing coresets of plans and extend this notion to robust coresets, which incorporate uncertainty. We prove positive and negative results on the sizes of such coresets for common cost-function classes. Because our lower bounds suggest that the size of a coreset can be large in the worst case, we exploit the fact that in practice queries arise from workload distributions, rather than being arbitrary. We present algorithms that construct workload-aware coresets whose size and performance closely match those of the optimal workload-specific coresets.
We study linear sketches and data stream algorithms for the ℓp-sampling problem. Here one chooses an r x n matrix A and computes A · v for an underlying vector v. From A · v, one should output a coordinate i ∈ {1, 2, . . . , n} with probability (1 ± ε) ∥vi∥p/ ∥v ∥ p/ p . One is also allowed to output FAIL with a small constant probability. In the estimation version of the problem, one would also like to output a (1 ± ε)-approximation to ∥vi∥p/ ∥v ∥ p/ p, which is the probability for which i was sampled. For 0 < p < 2, work of Jayaram and Woodruff (FOCS, 2018) showed an upper bound of r = O(logn) while for p = 2 the best known upper bound is r = O(log 2 n). They posed it as an open question to resolve this gap. We prove an Ω(log 2 n) sketching dimension lower bound for p = 2, thereby separating the complexity for 0