The advancements in Location Based Services (LBS) and Location Based Social Networks (LBSN) have spurred multiple research efforts in query processing as well as recommendation systems that enable planning trips based on combining location and semantic properties of Points of Interest (POI). However, often times such trips need to involve the reality of existing road networks, for the purpose of obeying constraints such as distance or travel-time. Although there are many publicly available datasets (e.g., Gowalla) that include check-in data at POIs with location, they are often not integrated with existing roads-based data (e.g., Open Street Maps (OSM)) causing researchers to spend extra time and labour to experimentally evaluate their findings. In this paper, we present: (1) methodologies for extracting information regarding POIs from publicly available datasets based on users posting; (2) extracting concise semantic categories for each POI; (3) integrating their location and semantic categories with an existing road network. In addition to the methodologies, we also provide two datasets (based on POIs and road networks in Chicago and New York City) constructed using our methodologies that researchers can readily use for their semantic-aware POIs with location and trip based query processing tasks as well as deep learning tasks.

One of the most popular applications of Location-Based Services (LBS) is to recommend proximal Points of Interest (PoIs) – e.g., nearby restaurants and museums, police stations, hospitals, etc. – or a sequence of PoIs to visit. An important variant of the problem that is recently addressed not only considers the preference of distance/proximity, but also desires that the returned proximal objects satisfy certain semantic constraints. For instance, rather than picking several close-by attractions with similar features – e.g., restaurants with similar menus; museums with similar art exhibitions – a tourist may be more interested in a result set that could potentially provide more diverse types of experiences, as long as they are within an acceptable distance from a given location. In my PhD dissertation, a topic modeling approach based on the Latent Dirichlet Allocation (LDA) was introduced to effectively model and exploit a fine-grained notion of diversity, based on sets of keywords and/or reviews instead of a coarser user-given category. More importantly, for efficiency purpose, we devised two novel indexing structures – Diversity Map and Diversity Aggregated R-tree, which in turn enabled us to develop efficient algorithms to generate the answer-set for two novel categories of queries. While both queries focus on determining the recommended locations among a set of PoIs that maximizes the semantic diversity within distance limits along a given road network, they each tackle a different variant. The first type of query is kDRQ, which finds k such PoIs with respect to a given user’s location. The second query kDPQ generates a path to be used to visit a sequence of k such locations (i.e., with max diversity), starting at the user’s current location, and its extended variant kTDPQ allows user to specify not only starting location but also destination. Moreover, we tackled another query PaDOC of jointly generating a path that: (1) can be traversed within a user-specified budget (e.g., limit on distance); (2) starts at one of the user-specified origin locations (e.g., a hotel); and (3) contains PoIs from a user-specified list of PoI categories. Based on a novel indexing structure k-Closest Category Matrix, we proposed two efficient algorithms for approximate PaDOC query processing based on both conservative and progressive distance estimations.

Understanding human mobility has become an important aspect of location-based services in tasks such as personalized recommendation and individual moving pattern recognition, enabled by the large volumes of data from geo-tagged social media (GTSM). Prior studies mainly focus on analyzing human historical footprints collected by GTSM and assuming the veracity of the data, which need not hold when some users are not willing to share their real footprints due to privacy concerns—thereby affecting reliability/authenticity. In this study, we address the problem of Inferring Real Mobility (IRMo) of users, from their unreliable historical traces. Tackling IRMo is a non-trivial task due to the: (1) sparsity of check-in data; (2) suspicious counterfeit check-in behaviors; and (3) unobserved dependencies in human trajectories. To address these issues, we develop a novel Graph-enhanced Attention model called IRMoGA , which attempts to capture underlying mobility patterns and check-in correlations by exploiting the unreliable spatio-temporal data. Specifically, we incorporate the attention mechanism (rather than solely relying on traditional recursive models) to understand the regularity of human mobility, while employing a graph neural network to understand the mutual interactions from human historical check-ins and leveraging prior knowledge to alleviate the inferring bias. Our experiments conducted on four real-world datasets demonstrate the superior performance of IRMoGA over several state-of-the-art baselines, e.g., up to 39.16% improvement regarding the Recall score on Foursquare.

Route planning based on user’s preferences and Points of Interests (POIs) is one of the most popular applications of Location-Based Services (LBS). Variants of route planning consider distance constraints (e.g., the maximum length of the route), origin constraints (e.g., a set of possible starting locations of the route), and category constraints (e.g., a multiset of POI categories that the route must visit). However, the problem of deciding whether a route exists that visits all required POI categories under the distance constraint is known to be NP-hard. Assuming P ≠ NP, this means that there is no efficient (polynomial time) solution to find such paths. Recently, approximate algorithms have been proposed for searching for such a path. This demonstration leverages several of these algorithms to provide a web-based system with a graphical user interface (UI) which allows the users to find a path that: (a) satisfies a distance limit; (b) generates a route to visit a list of POIs, based on the user’s preferred categories; (c) provides a set of hotels (as possible starting locations of the path). If the approximate search algorithms are able to find such a path, it will be displayed on a Mapbox-based map interface that shows: (1) all POIs on a path and (2) alternative paths if any were found. The system then allows a user to explore the returned paths, select a path, or refine their constraints. Moreover, the system allows the users to select which approximate algorithm they would prefer to execute.

Traditionally, clustering of multivariate data aims at grouping objects described with multiple heterogeneous attributes based on a suitable similarity (conversely, distance) function. One of the main challenges is due to the fact that it is not straightforward to directly apply mathematical operations (e.g., sum, average) to the feature values, as they stem from heterogeneous contexts.

One of the most popular applications of Location Based Services (LBS) is recommending a Point of Interest (POI) based on user's preferences and geo-locations. However, the existing approaches have not tackled the problem of jointly determining: (a) a sequence of POIs that can be traversed within certain budget (i.e., limit on distance) and simultaneously provide a high-enough diversity; and (b) recommend the best origin (i.e., the hotel) for a given user, so that the desired route of POIs can be traversed within the specified constraints. In this work, we take a first step towards identifying this new problem and formalizing it as a novel type of a query. Subsequently, we present naïve solutions and experimental observations over a real-life datasets, illustrating the trade-offs in terms of (dis)associating the initial location from the rest of the POIs.

We present CACSE – a system for Context Aware Clustering of Stellar Evolution – for datasets corresponding to temporal evolution of stars, which are multivariate time series, usually with a large number of attributes (e.g., ≥ 40). Typically, the datasets are obtained by simulation and are relatively large in size (5 ∼ 10 GB per certain interval of values for various initial conditions). Investigating common evolutionary trends in these datasets often depends on the context – i.e., not all the attributes are always of interest, and among the subset of the context-relevant attributes, some may have more impact than others. To enable such context-aware clustering, our CACSE system provides functionalities allowing the domain experts to dynamically select attributes that matter, and assign desired weights/priorities. Our system consists of a PostgreSQL database, Python-based middleware with RESTful and Django framework, and a web-based user interface as frontend. The user interface provides multiple interactive options, including selection of datasets and preferred attributes along with the corresponding weights. Subsequently, the users can select a time instant or a time range to visualize the formed clusters. Thus, CACSE enables a detection of changes in the the set of clusters (i.e., convoys) of stellar evolution tracks. Current version provides two of the most popular clustering algorithms – k-means and DBSCAN.

Geospatial data integration combines two or more data layers to facilitate advanced querying, analysis, reasoning, and visualization. In general, different layers (e.g., ZIP codes, census blocks, school districts, and land use parcels) have different spatial partitions and different types of associated semantic descriptors. In addition, geospatial data may contain errors (e.g., due to imprecision in the measurements or to representation constraints) causing uncertainty that needs to be incorporated and quantified in the query answers. In this paper, we leverage semantic descriptors in heterogeneous information layers to build a data structure that enables efficient processing of geospatial range queries by returning an estimate of the answer together with an error bound. We present the processing algorithms and evaluate our approach by means of experiments that encompass large datasets, demonstrating the benefits of our approach.

Location-Based Services are often used to find proximal Points of Interest PoI - e.g., nearby restaurants and museums, police stations, hospitals, etc. - in a plethora of applications. An important recently addressed variant of the problem not only considers the distance/proximity aspect, but also desires semantically diverse locations in the answer-set. For instance, rather than picking several close-by attractions with similar features - e.g., restaurants with similar menus; museums with similar art exhibitions - a tourist may be more interested in a result set that could potentially provide more diverse types of experiences, for as long as they are within an acceptable distance from a given (current) location. Towards that goal, in this work we propose a novel approach to efficiently retrieve a path that will maximize the semantic diversity of the visited PoIs that are within distance limits along a given road network. We introduce a novel indexing structure - the Diversity Aggregated R-tree, based on which we devise efficient algorithms to generate the answer-set - i.e., the recommended locations among a set of given PoIs - relying on a greedy search strategy. Our experimental evaluations conducted on real datasets demonstrate the benefits of proposed methodology over the baseline alternative approaches.

Proximity-oriented spatial queries, such as range queries and k-nearest neighbors (kNNs), are common in many applications, notably in Location Based Services (LBS). However, in many settings, users may also desire that the returned proximal objects exhibit (likely) maximal and fine-grained semantic diversity. For instance, nearby restaurants with different menu items are more interesting than close ones offering similar menus. Towards that goal, we propose a topic modeling approach based on the Latent Dirichlet Allocation, a generative statistical model, to effectively model and exploit a fine-grained notion of diversity, namely based on sets of keywords (e.g., menu items) instead of a coarser user-given category (e.g., a restaurant's cuisine). In addition, and relying on the notion of Distance Signatures, we propose an index structure that can be used to effectively extract the k objects that are within a range distance from a given query location, and which are also semantically diverse. Our experimental evaluations using real datasets demonstrate that the proposed methodology is able to provide highly diversified answers to cardinality-wise constrained range queries much more efficiently than a straightforward alternative solution.