TL;DR
This paper introduces sMash, a system designed to simplify the creation of web data mashups. It tackles the complexity of discovering and combining data APIs by offering a visualized API network, intuitive "surfing" capabilities, and intelligent recommendations, making mashup creation more accessible and inspiring for a wider range of users.
What Problem Does This Paper Address?
The proliferation of web data APIs has opened up immense possibilities for creating new applications by combining existing services (mashups). However, building these mashups is often challenging for users, especially those new to the field. The paper identifies three key problems:
- Discovery and Understanding: Users struggle to find the right data APIs and understand how they relate to each other.
- Complexity of Integration: The process of integrating APIs, from reading specifications to deciphering functions and checking compatibility, is cumbersome.
- Limited Discovery: Users often lack the ability to discover potentially useful APIs that could enrich their mashups. Existing tools, while helpful, don't fully address these issues, particularly in providing an intuitive way to explore and combine APIs.
Methods
sMash employs a multi-faceted approach to address these challenges:
- Semantic-based Mashup Graph: sMash automatically constructs a network of mashupable APIs. This graph is visualized, allowing users to "surf" through APIs like web pages.
- Fuzzy Keyword Search: For users with a clear idea of desired APIs, sMash offers a "fuzzy-match-keyword-search" that automatically generates a mashup graph.
- Trace-based Navigation and Recommendation: Users' interactions (traces) within the API network are recorded. This data is used to provide real-time recommendations for the next most popular links to explore, helping novice users navigate.
- Inference-based Recommendation: This feature targets power users by analyzing the existing mashup repository. It identifies and recommends underutilized APIs and links that might offer richer mashup possibilities, aiming to uncover hidden potential.
- Network Construction: The system uses RDF for precise API metadata representation, incorporating semantics, tags, categories, and data content. It also leverages user-contributed schemas and data types for scalability.
- Automatic Parameter Mapping: sMash automates the mapping of output parameters from one API to input parameters of another, simplifying the connection process.
Datasets
The paper mentions using a "repository" formed by extracting mashups from the communities of Yahoo Pipes and ProgrammableWeb.com for evaluation and inference-based recommendation. No new datasets were introduced.
Key Insights and Ideas
- Visualizing the API Landscape: Presenting data APIs as a navigable network, akin to the World Wide Web, significantly lowers the barrier to entry for mashup creation.
- Leveraging User Traces: User interaction data is a powerful signal for guiding novice users and suggesting relevant next steps, making the exploration process more efficient.
- Uncovering Hidden API Potential: A significant portion of available APIs are underutilized. By analyzing existing mashups and recommending less obvious connections, sMash can inspire more innovative and abundant mashups.
- Bridging the Gap Between Novice and Power Users: The system offers distinct features (trace-based vs. inference-based recommendations) tailored to different user expertise levels.
Key Results
The paper demonstrates that sMash can maintain high-quality recommendations with reasonable processing times, even with growing repository sizes. The evaluation of inference-based recommendation shows that it can suggest approximately 50% new, relevant links, significantly expanding the creative space for users. The system's interface is designed for intuitive interaction, allowing users to easily master APIs and build mashups.
Limitations and Future Work
- Limitations:
- The paper acknowledges that a "little pruning" might be needed to remove redundant links in automatically generated graphs.
- While the system aims to be comprehensive, the scalability of the network construction relies on user contributions, which could be a bottleneck.
- The evaluation of inference-based recommendation is based on a simulated scenario where a portion of existing mashups are used as input.
- Future Work:
- The authors express a desire to discuss with other participants how visualization and recommendation could be further improved.
- They also aim to explore additional features to refine the design and implementation of sMash.
Summary updated