Cite4Me: A Semantic Search and Retrieval Web Application for Scientific Publications

Bernardo Pereira Nunes, Besnik Fetahu, Stefan Dietze, Marco A. CasanovaInternational Semantic Web Conference (Posters & Demos) 2013 · OpenReview

Introduction

The rapid expansion of academic literature has made identifying relevant research increasingly difficult. Traditional digital libraries often rely on keyword-based searches or rigid hierarchical classifications, which frequently fail to capture the conceptual nuances of a field. Cite4Me addresses this by introducing a "semantic layer" over academic metadata, utilizing Linked Open Data (LOD) and graph-based algorithms to facilitate deeper exploration of scientific publications beyond simple string matching.

Standard search engines are often "semantically poor," relying on exact nomenclature. Users who are unfamiliar with specific sub-field terminology may miss critical papers. Furthermore, these systems rarely illustrate the "lateral" connections between different disciplines. Cite4Me aims to bridge this gap by moving from simple keyword matching to concept-based discovery, helping researchers unveil new nomenclature and explore topic coverage more efficiently.

Methodology

Cite4Me enriches standard bibliographic data through several key processes designed to transform flat text into a structured knowledge graph.

Entity Extraction and Annotation

The system utilizes the DBpedia Spotlight API to scan paper abstracts and identify specific entities, types, and categories. This process transforms unstructured text descriptions into a set of standardized concepts linked to the DBpedia knowledge base. By mapping text to the dcterms:subject property in DBpedia, the system creates a common language for disparate papers.

Semantic Search and Dynamic Context

The search functionality is inspired by Explicit Semantic Analysis (ESA). It calculates relatedness scores between user queries and publications based on the $tf-idf$ (term frequency-inverse document frequency) of the entities found within the text, rather than just the raw words. To provide immediate visual feedback, the interface includes a dynamic tag cloud that updates as a user browses, reflecting the thematic flavor of search results or specific paper clusters.

Cite4Me search interface featuring a dynamic tag cloud and semantic search results.

Graph-Based Recommendation

Recommendations are generated using a variation of the Katz index, a measure from Social Network Theory. This index calculates the connectivity weight between concepts within the DBpedia knowledge graph. By aggregating these weights, the system can recommend papers that are conceptually related even if they share no common keywords, effectively finding "lateral" connections across different research domains.

Implementation and Results

The application currently operates on the Learning Analytics and Knowledge (LAK) dataset, which contains 315 full papers and their descriptive metadata from the ACM Digital Library. The metadata and full text were converted into RDF (Resource Description Framework) and made accessible via a SPARQL endpoint, interlinked with DBpedia.

Visual Discovery

One of the primary features of Cite4Me is its exploratory graph search. Instead of a standard list of links, publications are visualized as nodes in a graph. Links are created between papers that share DBpedia categories, allowing users to navigate research clusters visually.

Exploratory graph search interface showing research clusters and interlinked publications.

The system successfully demonstrates a pipeline for mapping unstructured academic text to the LOD cloud. Qualitative results show that the "Exploratory Search" interface effectively visualizes topically related publications as distinct clusters, providing a "map" of the research field.

Discussion and Future Work

While Cite4Me offers a novel approach to literature discovery, it currently faces limitations regarding dataset scale and copyright restrictions. Expanding the system to handle millions of documents and navigating the legal hurdles of full-text access remain significant challenges. Future work will focus on formal user evaluations, such as A/B testing against standard keyword search, and the integration of citation sentiment analysis to further refine the retrieval process.

Summary updated

BibTeX
@inproceedings{nunes2013cite,
  author = {Bernardo Pereira Nunes and Besnik Fetahu and Stefan Dietze and Marco A. Casanova},
  title = {{Cite4Me: A Semantic Search and Retrieval Web Application for Scientific Publications}},
  booktitle = {International Semantic Web Conference (Posters \& Demos)},
  year = {2013},
  url = {https://openreview.net/forum?id=Idxv9vsNaNd},
}
Figure