TL;DR
This paper introduces RDF EXPLORER, a visual interface and language designed to make querying large knowledge graphs like Wikidata more accessible to non-experts. A usability study demonstrated that users successfully completed more tasks and formulated more accurate query graphs using RDF EXPLORER compared to the existing Wikidata Query Helper.
What Problem Does This Paper Address?
Knowledge graphs (KGs) are becoming increasingly prevalent, storing vast amounts of structured information. However, accessing and querying this data often requires specialized knowledge of query languages like SPARQL, which can be a significant barrier for non-expert users. Existing systems that aim to simplify this process often face a trade-off between expressivity (ability to form complex queries) and usability. Some systems support only simple, acyclic queries, while others assume a level of technical expertise. This paper addresses the challenge of enabling non-experts to effectively explore and query complex, cyclic graph patterns within KGs without requiring them to learn formal query languages.
Methods
The core of RDF EXPLORER is its visual query language (VQL) and the accompanying interface. VQL allows users to construct queries by visually building a "visual query graph" (VQG). This VQG is a directed, edge-labeled graph where nodes can be IRIs, literals, or variables. The language is built upon four atomic operations: adding a variable node, adding a constant node, adding an edge with a variable label, and adding an edge with an IRI label. These operations allow users to incrementally build query patterns, including cyclic ones.
The RDF EXPLORER interface translates these VQGs into SPARQL queries. Key features of the interface include:
- A visual query editor for building VQGs.
- A search pane for finding entities and properties, with auto-completion.
- Drag-and-drop functionality for adding nodes and edges.
- Suggestions for IRI labels to avoid empty results.
- Result previews for variables, allowing users to explore the graph interactively.
- The ability to convert constant nodes to variables, facilitating generalization from specific examples to graph patterns.
Datasets
The evaluation of RDF EXPLORER used Wikidata as the knowledge graph. Wikidata is a large, multilingual knowledge base containing information on tens of millions of items. No new benchmarks were introduced; the study focused on comparing the usability of RDF EXPLORER against an existing tool for Wikidata.
Key Insights and Ideas
- Visualizing Query Structure is Key: The study suggests that users struggle more with formulating the correct shape of a query graph when using a form-based interface (like the baseline) compared to a visual, graph-based approach. Conversely, users find it easier to identify correct terms (constants) with the form-based interface.
- Cyclic Queries are a Challenge: While RDF EXPLORER supports cyclic queries, the study indicates that users still face challenges with more complex query structures, suggesting that further improvements are needed for advanced query patterns.
- Interactive Feedback Enhances Usability: The result previews and suggestions provided by RDF EXPLORER seem to guide users effectively, helping them explore the graph and refine their queries.
- Trade-off Between Expressivity and Usability: The paper implicitly highlights the ongoing challenge of balancing the power of query languages with the ease of use for non-expert users. RDF EXPLORER appears to shift this balance towards usability for graph structure.
Key Results
The usability study involved 28 students and compared RDF EXPLORER against the Wikidata Query Helper (WQH). The results were analyzed at three levels of granularity: queries, edges, and shapes.
- Query Graphs: RDF EXPLORER significantly outperformed WQH in correctly formulating query graphs (H1 validated, t*=3.22).
- Edges: The difference in correctly formulating edges was not statistically significant (H2 not validated, t*=1.1947).
- Shapes: RDF EXPLORER also significantly outperformed WQH in generating correctly structured query graphs (H3 validated, t*=3.6431).
Qualitatively, users generally preferred RDF EXPLORER across most subjective criteria, with the exception of perceived physical effort (due to mouse usage) and lag in suggestion generation.
Limitations and Future Work
The authors acknowledge several limitations and suggest future directions:
- Complexity of Queries: The success rate for later tasks decreased, indicating that users still struggle with expressing more complex queries.
- Expressive Features: Future work could involve supporting more expressive SPARQL features like aggregations and unions.
- Performance: The system generates some lag when generating suggestions, which could be improved through approximation techniques.
- Generalizability: While tested on Wikidata, further evaluation on other knowledge graphs would be beneficial.
- Beyond SPARQL: The underlying language and interface could potentially be adapted to other graph query languages and models.
Summary updated