Semantic place retrieval is a popular research problem to search over knowledge graphs using both text and location information. While handling such queries, it is crucial to appropriately balance the relevance and spatial distance of places to the user's query, to satisfy the user's information needs. Furthermore, given modern users' expectations, it is also critical to returning results in a short time, which implies the necessity of using advanced data structures in the underlying retrieval system.
In this work, our contribution toward improving the efficiency of semantic place retrieval is two-fold. First, we show that by applying some ad hoc yet intuitive restrictions on the depth of search on the knowledge graph, it is possible to adopt several well-known data structures, so-called geo-textual indices that are introduced for processing the spatial keyword queries, for the semantic place retrieval scenario. Secondly, as a novel solution to the semantic place retrieval problem, we adapt the idea of cluster-skipping inverted index (CS-IIS), which has been originally proposed for retrieval over topically clustered document collections. In our adaptation, we also use an early-stopping technique based on the textual and spatial scores of the spatial grids that are being processed.
Our exhaustive experiments lead to several interesting findings. We show that while some of the earlier geo-textual indices in the literature yield high efficiency in terms of in-memory processing time, they may cause a large number of direct disk accesses. In contrast, our approach based on CS-IIS requires a few direct disk accesses (which is equal to the number of terms in the query) and hence, performs comparably or even better than the baseline approaches in terms of the total query processing time.