Effective and efficient retrieval of structured entities

Abstract

Structured entities are commonly abstracted, such as from XML, RDF or hidden-web databases. Direct retrieval of various structured entities is highly demanded in data lakes, e.g., given a JSON object, to find the XML entities that denote the same real-world object. Existing approaches on evaluating structured entity similarity emphasize too much the structural inconsistency. Indeed, entities from heterogeneous sources could have very distinct structures, owing to various information representation conventions. We argue that the retrieval could be more tolerant to structural differences and focus more on the contents of the entities. In this paper, we first identify the unique challenge of parent-child (containment) relationships among structured entities, which unfortunately prevent the retrieval of proper entities (returning parents or children). To solve the problem, a novel hierarchy smooth function is proposed to combine the term scores in different nodes of a structured entity. Entities sharing the same structure, namely an entity family, are employed to learn the coefficient in aggregating the scores, and thus distinguish/prune the parent or child entities. Remarkably, the proposed method could cooperate with both the bag-of-words (BOW) and word embedding models, successful in retrieving unstructured documents, for querying structured entities. Extensive experiments on real datasets demonstrate that our proposal is effective and efficient.

Approximate Event Pattern Matching over Heterogeneous and Dirty Sources

“…After getting the schema graph and the pattern graph, we try to find the most similar sub-structure of the schema graph with the pattern graph, and a target trace family corresponding to this sub-structure. The concept of trace family follows the idea of entity family [2], and is defined as the set of all the traces sharing a common sub-structure, which means all the events happen as the sub-structure defined in the traces. The similarity between any sub-structure S of the schema graph and the pattern graph p is computed as follows:…”

Schema Extraction on Semi-structured Data

“…Extracting tree/graph structured schema is important in query processing [3] and data repairing [4,5], etc. For example, entities from heterogeneous sources could have very distinct structures, owing to various information representation conventions.…”

“…Entity family could be regarded as a schema that defines the structure of the entities. [3] proposed a novel hierarchy smooth function to combine the term scores in different nodes of structured entities in the same entity family. Such an entity family is not only important in learning the parameters for effective score combination but also useful in pruning entities for efficient entity retrieval.…”

“…Informally, a match of the schema graph is a subgraph of the instance graph G to that each edge in the schema exactly corresponds to an edge in the subgraph, i.e., it is not required that each edge in G could be mapped to the schema graph. The graph repair algorithm only increases the evaluation of satisfaction on the graph matching problem [3], data repairing [4,5].…”

Against the Holistic Temptations

“…In recent years, embedding techniques are widely applied in various fields in database community, such as cardinality estimation [2,3], query optimization [4,5], language understanding [6], entity resolution [7,8], document retrieval [9], graph learning [10,11], and advertising recommendation [12], to learn the semantic representations of categorical features. Among these fields, Deep Learning Recommendation Models (DLRMs) are one of the most important applications of embedding techniques: they account for 35% of Amazon's revenue in 2018 [13][14][15], and consume more than 50% training and 80% inference cycles at Meta's data centers in 2020 [16,17].…”