TL;DR
The Web of Things (WoT) is generating a massive, heterogeneous stream of data from billions of interconnected devices. This paper provides a comprehensive taxonomy and survey of existing search techniques for this data, classifying them by fundamental principles, data representation, and content. It aims to offer a clear overview of the research landscape and identify future directions for efficient and effective data discovery in the WoT.
What Problem Does This Paper Address?
The Web of Things (WoT) paradigm aims to make physical world objects and their data accessible via standard web technologies, enabling intelligent applications and sophisticated data analytics. However, the sheer volume and heterogeneity of data generated by billions of connected devices pose a significant challenge. Traditional search techniques, designed for static web documents, are insufficient due to factors like device mobility, opportunistic sensing, continuous data streams with changing properties, and the need for efficient indexing of both historical and real-time data. This paper addresses the critical need for effective search mechanisms to navigate this complex data landscape and extract relevant information for WoT applications.
Methods
The paper's core contribution is a comprehensive taxonomy and survey of existing search techniques for the Web of Things. This taxonomy classifies techniques from three key perspectives:
- Fundamental Search Principles: This category focuses on the underlying techniques used for searching, broadly divided into indexing (e.g., text indexing, spatial indexing) and clustering (e.g., location-based clustering, non-location-based clustering).
- Data/Knowledge Representation: This perspective examines how data is structured and represented to facilitate search, with a particular emphasis on Linked Data and semantic streaming data.
- Content Being Searched: This viewpoint categorizes search based on what is being sought, including Observation and Measurement (O&M) data, sensor information, and entity information.
The authors review and categorize numerous existing research works based on this taxonomy, analyzing their strengths, weaknesses, and applicability to different WoT scenarios. They also discuss metrics for evaluating search techniques, such as data format, access approach, search type, scale of experiments, dynamicity, and architecture.
Datasets
The paper does not introduce new datasets. Instead, it surveys existing research that has utilized various datasets for evaluation. The specific datasets used by individual research works are not detailed in a consolidated manner, but the paper notes that "most of the works are performed in a closed environment and the datasets are not publicly available."
Key Insights and Ideas
- WoT Search is Fundamentally Different: The inherent dynamicity, mobility, and spatio-temporal nature of WoT data make it distinct from traditional web document search, requiring specialized approaches.
- No One-Size-Fits-All: The diverse requirements of WoT applications (e.g., smart cities, home automation, manufacturing) necessitate a variety of search techniques, with no single method being universally optimal.
- Semantics are Crucial: Leveraging semantic web technologies like Linked Data and ontologies is vital for enabling interoperability and richer knowledge representation, which in turn enhances search capabilities.
- The "Big Data" Challenge is Real: The overwhelming volume, velocity, and variety of WoT data necessitate a shift towards big data processing paradigms and distributed intelligence for effective search.
- Evaluation is Difficult: A significant limitation highlighted is the lack of standardized benchmarks and publicly available datasets, making direct comparison and evaluation of different search techniques challenging.
Key Results
The paper provides a detailed comparative analysis of numerous search techniques across the defined taxonomy. While not presenting new experimental results, it synthesizes findings from existing research, highlighting:
- Prevalence of Java: Most surveyed systems are implemented in Java, likely due to the availability of Java-based open-source tools and APIs.
- Dominance of REST and SPARQL: HTTP REST and SPARQL are identified as popular access methods due to their simplicity.
- Dynamicity is Key: A significant portion of research considers dynamicity in their designs, acknowledging its critical impact on WoT search.
- Convergence in Content Search: Search techniques for O&M data, sensors, and entities are converging, especially for applications dealing with physical world data.
Limitations and Future Work
Limitations:
- Lack of Standardized Benchmarks: The paper explicitly states that direct comparison of techniques is difficult due to the unavailability of benchmarks that can handle all types of search techniques.
- Closed Environments and Private Datasets: Most research is conducted in closed environments with non-public datasets, hindering reproducibility and broader evaluation.
- Limited Scalability of Some Techniques: Certain indexing and clustering approaches face scalability issues with large numbers of objects or data points.
- Difficulty in Embedding Rich Semantics: Existing web standards, while useful for discovery, may struggle to embed complex descriptions and rich semantics into WoT objects.
Future Work:
- Big Data Search: Developing more effective and efficient techniques for searching the massive WoT data, considering volume, velocity, and variety, and potentially redesigning existing building blocks to leverage big data technologies.
- Distributed Intelligence: Implementing search functionalities on distributed, autonomous units that cooperate to provide transparent services, hiding complexity and handling dynamic factors locally.
- Information Abstraction: Supporting search for abstract data objects that encode richer information through aggregation, approximation, and inference techniques.
- Personalisation: Moving beyond intelligence-centric approaches to develop more personalized search mechanisms that consider spatial, temporal, and individual preferences, enabling automated adaptation of search results.
- Comprehensive Evaluation: The need for standardized evaluation methods and publicly available datasets to facilitate robust comparisons of search techniques.
Summary updated